Every Seat in the Book

Requires

Your book's rough composition: classes, revenue bands, retention, and which accounts you actually enjoy servicing. Use counts and categories, not client names.

Steps

1. Prompt: "Here is my book by class, size band, revenue and retention: [paste anonymized]. Where do I make money per hour of service, and where do I only look busy? Show your reasoning."
2. Then: "Draft three candidate ICPs from this: class, size band, geography, buying trigger. For each: why I'd win it, what I'd have to get good at, and what I'd have to say no to."
3. Then: "Write the one-paragraph territory definition I could tape above my desk."

Guardrails

The AI argues from the numbers given, not from industry folklore about which classes are lucrative. No client names or identifying details go in: bands and counts carry the analysis.

Hand-back

The say-no list is the real product: an ICP you won't enforce is a poster. Pick one, and let the next ninety days test it before revisiting.

Requires

Your current opportunity list: company, class, stage, next step, date of last touch. Company names only, no individual contact details.

Steps

1. Prompt: "Here is my pipeline: [paste]. Define clean stage criteria for a commercial insurance sale (identified, contacted, meeting held, submission out, quoted, closing) and re-stage every entry against them. Flag entries with no next step or no touch in 21+ days."
2. Then: "Which three opportunities does the data say I'm neglecting relative to their stage and size?"
3. Weekly: paste the updated list with "same rules: what moved, what stalled, what should I do first this week?"

Guardrails

The AI restages on evidence in the row, never optimism: an opportunity with no next step is stalled by definition. Individual names, emails, and phone numbers stay out of a general assistant: company names are enough to run the review.

Hand-back

The weekly ten minutes is the skill: the AI keeps the list honest, you make the calls. Anything flagged twice in a row gets a decision, not a third flag.

Requires

Whatever fragments you have: business cards, call notes, "we renew in the fall" comments. Company names and dates, no personal contact details.

Steps

1. Prompt: "Build me an x-date tracker structure for commercial prospects: company, class, carrier if known, x-date or best guess, source of the date, confidence, when to start outreach (90 to 120 days out), and last touch. Then load these fragments into it: [paste]."
2. Then: "Which entries have guessed dates I should verify, and what's a natural, non-pushy way to ask for the real date in each situation?"
3. Monthly: "Here's the updated list. What enters the 120-day window in the next 60 days?"

Guardrails

A guessed date is marked as a guess and never silently hardens into a fact. The asking-scripts are conversation starters, not scripts to read: instruct plain, brokerly wording.

Hand-back

The list is only as good as the capture habit: every conversation where a renewal month gets mentioned earns a row that same day. The 120-day alarm is the payoff; work it.

Requires

Ten minutes after any decision, won or lost. Describe the account by class and size, not name.

Steps

1. Prompt: "Interview me about a deal I just [won/lost]: ask one question at a time about how it was sourced, who else was in it, what the incumbent did, what price did and didn't decide, and what the buyer said. Then write the post-mortem: what actually decided this, in one paragraph, plus one thing to repeat or change."
2. Quarterly: paste the collected post-mortems. "What patterns show up across these? What am I telling myself that the record doesn't support?"

Guardrails

The AI pushes past the first answer: "they went with a cheaper quote" gets a follow-up, because price is the reason buyers give, not always the reason they act. No flattery in the quarterly read: instruct it to name the uncomfortable pattern if one exists.

Hand-back

One repeat-or-change item per deal, actually applied to the next pitch. The quarterly pattern read is where the compounding happens; calendar it.

Requires

The company name and anything public: their website, news, directory listings.

Steps

1. Prompt: "Research brief on [company] from this material: [paste site text/links' content]. What do they do, for whom, at what apparent scale? Locations, operations, anything that changed recently (growth, new lines, leadership). Mark every inference as an inference."
2. Then: "What three things about this operation would a commercial broker want to ask about first, and why?"

Guardrails

Facts and inferences stay separated: 'appears to' is required wording for anything not stated in the source. No filling gaps with what similar companies usually do unless labeled exactly that.

Hand-back

Verify anything you'll say out loud on the call: quoting a wrong fact about their own business ends credibility in one sentence. The three questions are your discovery opener.

Requires

The research brief from Company and operations research, or the same public material.

Steps

1. Prompt: "From this public information about [company], build an exposure pre-read: the risks inherent to this operation, organized by line (property, liability, auto, workers comp, professional, cyber, umbrella). For each: why this operation has it and what would make it bigger or smaller. Label everything as a pre-read from public info, to be confirmed."
2. Then: "Which two exposures here are most likely to be poorly handled in a typical program for this class? Explain, without asserting anything about their actual coverage."

Guardrails

The AI describes exposure, never their coverage: it has not seen their program and must not pretend to. No fear framing: this document may end up read aloud to the prospect, and it should sound like competence, not a scare.

Hand-back

This is your meeting prep, not a deliverable yet: the Risk Scan (Risk Scan for prospects) is the polished version. Walking in already understanding their risk is the entire point of the pre-read.

Requires

Whatever you've gathered: comments from the prospect, public filings, certificate sightings, industry chatter. Notes, not their documents.

Steps

1. Prompt: "Here's what I know about [company]'s current insurance arrangement: [paste fragments]. Organize into: confirmed (with source), probable, unknown. What does the pattern suggest about who holds it and when it renews?"
2. Then: "For each unknown, give me a natural way to learn it in conversation without sounding like I'm casing the account."

Guardrails

Confirmed means someone said it or a document showed it: pattern-matching stays in probable. The conversation lines must be honest questions a broker would actually ask, not manipulation scripts.

Hand-back

The x-date and the incumbent's name are the two facts that matter: everything else is texture. When you get them, they go straight into the x-date list (X-date list building), same day.

Requires

The company name and public professional information: their site's team page, LinkedIn profiles you can see, news mentions.

Steps

1. Prompt: "From this public material on [company], map who likely owns the insurance decision: title, why them, and who else influences it (CFO, owner, office manager, depending on size). Note where you're inferring from company size rather than evidence."
2. Then: "Given who I am (a commercial broker with [your angle]), what's the most credible path to this person: direct, referral through someone we might share, or via an event or association they're visibly part of?"

Guardrails

Public professional information only: no hunting for personal emails, home details, or anything a person didn't publish professionally. Inference from company size is labeled as such.

Hand-back

Verify the person still holds the role before the first touch: org pages go stale. The path recommendation is a suggestion; you know your market's etiquette better than any model does.

Requires

The research brief (Company and operations research) and your own voice rules: length, sign-off, what you never say.

Steps

1. Prompt: "Write a cold email to the [title] at [company]. Use exactly one specific, true observation from this research: [paste]. My rules: under 120 words, no buzzwords, no false urgency, one soft ask (a specific question or a 15-minute call), sounds like a person. Draft three versions with different openers."
2. Then: "Critique all three as the recipient: which would you delete, which would you answer, and why?"

Guardrails

One observation, truly specific: 'I noticed you're growing' is banned wording. No invented commonality, no fake familiarity, no urgency that doesn't exist. Compliance basics stay in: real name, real firm, and the recipient can tell you're selling.

Hand-back

Pick one, make one edit in your own hand (there's always one line only you would write), send. Log the send in the pipeline (Pipeline tracking) with a next-touch date before you move on.

Requires

Your ICP (ICP and territory definition), one true story from your production years, and the objections you actually hear.

Steps

1. Prompt: "Build me a cold-call skeleton: a ten-second opener that earns thirty more, a permission line, one value statement built on this true story: [paste], and turns for these objections: [paste yours: 'we have a guy', 'send me something', 'renewal's not for months']. Spoken language: read it aloud in your head before giving it to me."
2. Then: "Strip it to a one-page call sheet: opener, value line, three turns, and the two next-step asks (meeting or x-date)."
3. Before a call block: "Quiz me: throw objections at me one at a time and critique my answers."

Guardrails

Nothing on the sheet the broker wouldn't say in their own voice: the AI adapts to pasted phrasing, not sales-book diction. The turns respect a real no: persistence lines that ignore refusal are out.

Hand-back

The sheet is a floor, not a script: the rep who sounds like they're reading loses. The quiz rep before a block is the underused half of this skill.

Requires

The prospect's public profile and recent activity, plus your own last few posts for voice.

Steps

1. Prompt: "Draft a connection note (under 280 characters) to [name, title] referencing this real thing from their public activity: [paste]. No pitch in the note. Then draft the first DM for after they accept: one useful observation or question about their business, one line of who I am, no ask yet."
2. Then: "Now the second DM for two weeks later, assuming no reply: different angle, still no pressure, one soft ask."

Guardrails

Reference only what they posted publicly and recently: quoting something old or obscure reads as surveillance, not attention. No pitch before the relationship exists: the note that sells gets ignored, and worse, remembered.

Hand-back

Send at most a handful a day, each one read once by you first: this channel runs on judgment per person, and automation smell kills it. Log accepts and replies in the pipeline.

Requires

The first touch that was sent, and anything true that's happened since (their news, a relevant market change, a useful article).

Steps

1. Prompt: "Here's my first email to [company]: [paste]. Draft touches two through five over eight weeks. Each must carry something new (an insight for their class, a relevant change, a genuinely useful resource), get shorter as it goes, and never open with 'just following up' or 'bumping this'. Touch five is the polite close: door open, pressure zero."
2. When something real happens: "This just happened: [news]. Draft the touch it justifies, replacing whatever was next in the sequence."

Guardrails

No touch exists without a reason the recipient would recognize as value: cadence alone is not a reason. Guilt lines and fake breakups are banned. Every email is one the broker would be comfortable having forwarded to the prospect's boss.

Hand-back

Dates go in the calendar the day touch one sends: sequences die from forgetting, not from rejection. A reply at any point moves the account out of sequence and into a real conversation, immediately.

Requires

The moment: a win, a save, a smooth renewal, a thank-you from the client. Describe it without naming the client.

Steps

1. Prompt: "A client just [describe the win, no names]. Draft two versions of a referral ask for this moment: one for the end of a phone call (spoken, two sentences), one for email (three sentences). Specific about who I'd like to meet ([your ICP wording]), zero pressure, easy to decline gracefully."
2. Then: "Draft the thank-you I send when a referral actually arrives, and the update I send the referrer after I've met their contact, whether or not it went anywhere."

Guardrails

The ask names a type of introduction, never a specific person the AI invented. It must be declinable without awkwardness: pressure spends the relationship the win just built.

Hand-back

The AI drafts the words; the read on whether this client, this moment, can carry an ask is entirely yours. The follow-up loop (thank, then update) is what turns one referral into a referring habit.

Requires

The research brief (Company and operations research) and exposure pre-read (Exposure pre-read).

Steps

1. Prompt: "Build my discovery meeting plan for [company] from this research: an opening that shows I did the work without showing off, seven questions ordered from their business to their risk to their current arrangement, and the two things I must not leave without (decision process, x-date). One page."
2. Then: "For each question, what answer would make this account more attractive, and what answer is a red flag?"

Guardrails

Questions about their business come before questions about their insurance: brokers who lead with policy talk get treated like vendors. Nothing on the agenda assumes facts the research only inferred.

Hand-back

Memorize the opening and the two must-gets; hold the rest loosely. The red-flag answers are your qualification filter: hearing one is information, not failure.

Requires

The prospect's class and size band. This skill builds the blank intake; the completed one, full of their answers, is a client document and doesn't go back into a general AI.

Steps

1. Prompt: "Build a risk intake questionnaire for a [class, size] operation: sections for operations, property, people, vehicles, contracts, prior losses, current program. Questions a business owner can answer without a glossary, ordered so the easy ones come first. Flag which questions matter most for underwriting this class."
2. Then: "Cut it to the twenty questions that carry the most underwriting weight, for when I only get thirty minutes."

Guardrails

Plain language throughout: a question the owner has to ask you to explain is a failed question. Class-specific, not generic: the flagged-for-underwriting list is the test of whether the AI understood the class.

Hand-back

The completed questionnaire is where their confidential information starts: it lives in your file system, not in a chatbot. The twenty-question cut is usually the one you'll actually use.

Requires

The exposure pre-read (Exposure pre-read), your firm's letterhead conventions, and the discipline to keep it public-info only.

Steps

1. Prompt: "Turn this exposure pre-read into a two-page risk overview addressed to [company]: their operation as we understand it, the risk areas inherent to businesses like theirs, and three questions worth discussing with their broker, whoever that is. Professional, generous, zero pitch. Every statement flagged as based on public information."
2. Then: "Add the closing line: prepared by [you, firm], happy to walk through it, no strings."

Guardrails

Nothing in the document asserts anything about their current coverage: it describes risk, not their program. The no-pitch rule is absolute: the document is the pitch, precisely because it doesn't contain one.

Hand-back

Read it as the prospect's current broker would: if any line would embarrass you in that reading, cut it. This is the value-first move your whole outreach model is built on: send it, then be patient.

Requires

Your raw meeting notes. CLIENT FILE: this is what a prospect disclosed in confidence about their business. Strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the notes. Prompt: "Structure these meeting notes: facts learned about the operation, exposures and concerns they raised, their current arrangement as described, decisions made, promises made (by me and by them), open questions, next step with owner and date."
2. Then: "Draft the recap email: warm, brief, plays back what I heard in their words, confirms the next step, and lists exactly what I committed to send."

Guardrails

Promises made is the section that cannot be wrong: instruct the AI to list only commitments stated in the notes, verbatim where possible. The recap contains nothing the prospect didn't say: playing back an inference as their words breaks trust in one email.

Hand-back

Send the recap same day: speed here is half the impression. The structured notes go in the file: eighteen months from now, at the BOR moment, this document is your memory.

Requires

The prospect's current policies, handed to you for review. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the gap analysis skill (Coverage gap identification) pointed at someone else's placement: run it exactly, on their program and operation.
2. Then: "Separate the findings: genuine gaps and mismatches versus defensible judgment calls the incumbent may have made deliberately. Only the first list is a takeover argument."
3. Then: "Draft how I present the top three findings: factual, specific, zero disparagement of the current broker by name or implication."

Guardrails

Findings, not attacks: the presentation criticizes the program's state, never the person who built it, because the prospect chose that person and disparagement reads as an insult to their judgment. Defensible-call items stay out of the pitch.

Hand-back

Verify every finding you'll present against the documents: a takeover pitch with one wrong claim in it dies entirely. What you're really selling is the review itself: this is what looking closely feels like.

Requires

The situation, described without documents: how warm the prospect is, x-date, what the gap review found, market conditions for the class.

Steps

1. Prompt: "Talk me through a BOR decision. Situation: [describe]. Walk the options: BOR now, compete at renewal, or wait and nurture. For each: what has to be true for it to work, what the incumbent's counter looks like, what the prospect experiences, and where it can blow up."
2. Then: "Play the incumbent: I just got rescinded or countered. What did they likely say, and what's my move?"

Guardrails

The AI maps considerations and rehearses scenarios: it does not know your market's carriers, relationships, or this prospect's temperament, and must not pretend to. Timing rules of thumb are labeled as rules of thumb.

Hand-back

BOR timing is pure judgment with real relationship stakes: the decision is yours alone. What the rehearsal buys you is not the answer but the absence of surprise.

Requires

The insured's legal name, the carrier(s), policy number(s), and effective date. CLIENT FILE: this is exactly the identifying detail that stays out of a general AI. Use placeholders there, or run it where the file stays put.

Steps

1. Prompt: "Draft a broker-of-record letter with placeholders: [INSURED LEGAL NAME], [CARRIER], [POLICY NUMBERS], [EFFECTIVE DATE], appointing [your firm] as broker of record for the policies listed, on the insured's letterhead, for the insured's authorized signature. Standard rescission-period language. Nothing else in the letter."
2. Then: "List what I must confirm before sending: exact legal name matches the policy, every policy number, who is authorized to sign, and this carrier's BOR processing quirks if I know them."

Guardrails

Placeholders in the draft, real details only at your desk: the letter's power is precision, and its privacy risk is the same precision. No extra prose: a BOR letter that editorializes invites questions a clean one doesn't.

Hand-back

Fill the placeholders from the actual policies, not from memory: one digit off in a policy number stalls the whole move. Walk the client through what happens after they sign, including the incumbent's likely call to them.

Requires

Three of your past posts (for voice), and one real observation from your week: a coverage nuance, a market shift, a lesson.

Steps

1. Prompt: "Here are three posts of mine: [paste]. Match my voice exactly: sentence length, formatting habits, how I open and close. Now draft a post from this observation: [paste]. No hashtag walls, no engagement bait, no 'agree?', no hook formulas."
2. Then: "Give me two alternate openings: one that starts mid-story, one that starts with the counterintuitive point."

Guardrails

The observation stays true as told: no inflating a small lesson into a dramatic one. Client situations get anonymized past recognition or left out. Banned vocabulary: revolutionize, transform, disrupt, unlock, game-changer.

Hand-back

Edit one line into something only you would write; that line is why it reads human. Post at your cadence, not the algorithm's: showing up beats optimizing.

Requires

The topic you keep explaining on calls, your voice samples, and who reads this (clients, prospects, referral partners).

Steps

1. Prompt: "I explain [topic] to clients constantly. Interview me: five questions to pull out what I actually say, my examples, and where people get confused. Then draft the article from my answers, in my voice: [paste samples]. Written for [audience], readable in four minutes."
2. Then: "Cut anything a reader could get from a generic insurance blog. What's left should only be writable by someone who does this work."

Guardrails

The interview answers are the source: the AI adds structure, never industry filler. Coverage statements stay general; the article educates and never reads as advice about the reader's specific policy.

Hand-back

The cut test is yours to enforce: if a paragraph could appear on a carrier's blog, it goes. Publish on a schedule you'll keep for a year, not the one that sounds impressive.

Requires

The story as you'd tell it, anonymized: class, size band, the problem, what you did, the outcome. CLIENT FILE: the client's identity and specifics stay out of a general AI; names and confirmed details merge at your desk after written permission.

Steps

1. Prompt: "Draft a case study from this anonymized story: [paste]. Structure: the situation, what most brokers would have done, what we did, the outcome in the client's terms. Placeholders for name and any identifying figure. No hero framing: the client's decision is the win, we did the work."
2. Then: "Draft the permission email to the client: what we'd like to publish, exactly what would be named, and that nothing runs without their written OK."

Guardrails

Nothing identifiable enters the draft: if the class-plus-region combination points at one company, generalize further. Outcomes stated only as facts you can defend; a case study is a claim about your work.

Hand-back

Get the permission in writing, fill the placeholders, and show the client the final before it publishes: the relationship outranks the marketing asset every time.

Requires

Who they are professionally, how you know them, and something useful you could offer their clients (a checklist, an article, an answer pattern).

Steps

1. Prompt: "Draft a note to a [profession] I know from [context]. The offer: [the useful thing]. Frame: their clients face [insurance situation their clients actually face], here's something that helps, no referral ask in this note. Warm, three sentences, sounds like me: [paste voice sample]."
2. Then: "Draft the follow-up for a month later, after they've used it: the soft version of 'if a client ever needs a commercial broker's eyes, I'm easy to send them to'."

Guardrails

Value flows first and asks come later: a COI note that asks in message one becomes spam from a friend, which is worse than spam. The useful thing must be genuinely useful to their clients, not disguised marketing.

Hand-back

The relationship math is yours: which accountant, which lawyer, which lender actually serves your ICP. One good COI who sends two referrals a year outperforms a mailing list of fifty.

Requires

Your book's happy-client moments (renewal saves, smooth claims, gap catches) and what you can offer as thanks that fits your compliance rules.

Steps

1. Prompt: "Design a referral system for a commercial insurance broker: the trigger moments to ask (after which wins), the ask wording per moment (spoken and email versions), how the referred introduction gets handled so the referrer looks good, and the thank-you loop. Simple enough to run without software."
2. Then: "Draft the tracking sheet: who was asked, when, result, thank-you sent."

Guardrails

Compensation for referrals is regulated: the thank-you options must be flagged for a compliance check in your jurisdiction rather than assumed legal. Ask wording stays declinable without awkwardness, same rule as the referral-asks skill (Referral asks).

Hand-back

Pick two trigger moments and actually wire them into your workflow: a referral system is a habit wearing a document. Review the sheet quarterly; the pattern of who refers tells you who your real fans are.

Requires

Your book list with lines per client. CLIENT FILE: names plus coverage is client data; strip to anonymized IDs before a general AI, or run it where the file stays put.

Steps

1. Paste the anonymized book (ID, class, size band, lines held). Prompt: "For each account: which lines a business of this class and size typically carries that this account doesn't hold with us. Rank the gaps by how consequential the missing line is for that class, not by premium opportunity."
2. Then: "Draft the outreach frame per gap type: 'in reviewing accounts like yours' honesty, the risk the line addresses in one plain sentence, and an offer to look at it at renewal. Zero pressure wording."

Guardrails

The ranking is client-risk-first, premium-second, and the outreach says so honestly: cross-sell that reads as revenue harvesting spends trust. 'Typically carries' claims stay at class level, never asserting the client is exposed without reviewing their situation.

Hand-back

Re-attach real names at your desk and filter with what you know: which client just had a hard year, which one asked about this last spring. The list is a map; the judgment of who to call is the brokerage.

Requires

What you actually do and for whom (your ICP), your differentiators stated as facts, and any proof you're allowed to use.

Steps

1. Prompt: "Draft a one-pager for a commercial insurance broker: who we work with (specific), what working with us looks like (the process, plainly), what we're unusually good at (facts, not adjectives), and how to start. Every claim must be one I can defend in a meeting. No stock phrases: 'trusted advisor', 'peace of mind', 'one-stop shop' are banned."
2. Then: "Now the eight-slide version for when a deck is expected: same content, one idea per slide, speaker notes in my voice."

Guardrails

Claims are inventory, not aspiration: nothing goes in that isn't true today. Differentiators must differentiate; if every broker in town could say it, cut it.

Hand-back

The defend-in-a-meeting test is yours: read each line imagining a skeptical CFO asking 'how, specifically?' Update it twice a year; stale collateral quietly tells prospects the details don't matter to you.

Requires

Your best past proposal (anonymized) and the sections you always end up writing.

Steps

1. Prompt: "Turn this anonymized proposal into a reusable template: fixed sections (who we are, our process, disclosures), variable sections clearly marked ([SITUATION], [OPTIONS TABLE], [RECOMMENDATION]), and formatting rules so every producer's proposal looks like the same firm made it."
2. Then: "Write the usage note that goes at the top for whoever fills it: what never changes, what always gets rewritten fresh, and the sections where copy-paste is how errors happen."

Guardrails

Disclosure and coverage-governs language is fixed text, not variable: standardizing it is the E&O point of a template. The usage note names the copy-paste danger sections explicitly (limits, premiums, carrier names).

Hand-back

The client proposal skill (Client proposal) fills this template per deal: the template is the frame, never the thinking. Review it after every proposal that felt awkward; awkwardness is the template asking for an edit.

Requires

The happy moment, described without the client's name, and where reviews matter for you (Google, LinkedIn recommendations).

Steps

1. Prompt: "A client just [moment, anonymized]. Draft the review ask for right now: spoken version for the end of the call, email version with the direct link placeholder. Two sentences, specific about what feedback helps ('what it's like working with us'), easy to decline."
2. Then: "Draft the thank-you for after they post, and the graceful non-response if they never do."

Guardrails

No incentives for reviews and no drafting the review for them: both poison the asset and often the platform terms. The ask happens at the moment of the win, and the wording must never make declining awkward.

Hand-back

The timing read is yours, same as referral asks: this client, this moment. Reply to every review that appears, including the lukewarm one, in your own words.

Requires

Your current site and LinkedIn text, pasted, and what's changed since it was written.

Steps

1. Prompt: "Audit this bio and site copy against reality: [paste text, then paste what's true now]. List every line that's stale, vague, or says nothing ('we offer a wide range of solutions'), and draft the replacement for each in the same voice, tighter."
2. Then: "Rewrite the LinkedIn headline and about section for the person my outreach targets ([ICP]): what they'd need to see to reply to my cold email."

Guardrails

Claims audit rules apply: nothing on a public profile that isn't true today. The headline serves the prospect's question ('is this person worth replying to'), not vanity.

Hand-back

This seat is marked yours on the map for a reason: no shortcut replaces you deciding what your own name stands for. Calendar the audit twice a year; your cold email's reply rate is partly decided here.

Requires

The topic you can teach cold, the audience (association, COI's clients, prospects), and 45 minutes of their time.

Steps

1. Prompt: "Build a 30-minute talk outline on [topic] for [audience]: open with the expensive misconception, teach three things they can use Monday, close with questions. Slide-per-idea outline plus my speaking notes. No pitch anywhere in it: the talk is the pitch."
2. Then: "Draft the follow-up email to attendees: the promised materials, one extra resource, and a soft door ('happy to look at anything this raised')."

Guardrails

Teach real technique, not teaser content that requires hiring you to be useful: a room can smell withheld value. Coverage examples stay generic; no attendee should wonder if you're describing their policy.

Hand-back

Standing in front of the room is the job, and it's marked yours: the outline just means you walk in prepared. The follow-up sends within 24 hours; the door line is where the pipeline entries come from.

Requires

Your ICP (ICP and territory definition) and honesty about how much time you'll actually give this.

Steps

1. Prompt: "My ICP is [paste]. Where do these buyers actually gather: association types, chamber committees, industry events? For each venue type: what presence looks like at three effort levels (attend, contribute, lead), and what each level realistically produces in a year."
2. Then: "Given [X hours a month], build the plan: which one or two venues, at which level, and what 'working the room well' means for someone who'd rather be useful than schmooze."

Guardrails

The AI doesn't know your town: venue suggestions are types to verify locally, labeled as such. The plan fits the stated hours; an aspirational plan is a guilt generator.

Hand-back

No software replaces a handshake, which is why the map marks this yours. Pick the one venue, commit for a year, and volunteer for the job nobody wants: that's how outsiders become fixtures.

Requires

Your discovery notes and intake answers. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the notes. Prompt: "Map this operation: for each activity described, the ways it can go wrong (property, injury, liability to others, income interruption, contract exposure), stated plainly. Then flag activities described too thinly to map, as questions for the client."
2. Then: "Order the map by what would actually threaten this business's survival versus what would merely cost money."

Guardrails

The map covers only what the notes describe: no importing a generic risk list for the class and pretending it's this client. Survival-versus-cost ordering must be argued from their described scale, not asserted.

Hand-back

Walk the map with the client and let them correct it: the corrections are the best discovery you'll do. The thin-spots list is your next meeting's agenda.

Requires

The prior schedules and whatever documents show changes (new leases, purchase records, the client's list). CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste prior schedule and change documents. Prompt: "Update this schedule from these documents: every row, with CHANGED / UNCHANGED / NEW / REMOVED status and the source for each change. Anything the documents imply but don't confirm: FLAGGED, with the question to ask."
2. Then: "Draft the client confirmation email: the changes in plain language, the flagged questions, and the line that values and lists are theirs to confirm."

Guardrails

Sourced or flagged, never bridged: a schedule row is a representation that prices coverage and pays claims. Valuation questions (replacement cost, business income figures) are always the client's numbers to confirm, and the email says so.

Hand-back

Chase the flags before renewal season, not during: a stale SOV found in November is a schedule; found at a claim it's a coinsurance penalty. File the client's confirmation with the schedule.

Requires

The client's financials and their answers on recovery time. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the financials. Prompt: "Work the business income calculation from these statements: the revenue base, continuing versus non-continuing expenses categorized with reasoning shown, and the resulting exposure at 6, 12, and 18-month recovery periods. Every figure traced to its line in the financials; every categorization judgment flagged as a judgment."
2. Then: "List the questions only the client can answer: realistic rebuild time, whether customers wait or leave, seasonal peaks, extra expense to keep operating."

Guardrails

Categorizations (does payroll continue?) are judgments and must be shown as such, never buried in the math: those judgments are where BI limits go wrong. The AI checks arithmetic; it does not decide recovery time.

Hand-back

The client-questions list is a sit-down conversation, not an email: recovery assumptions are the whole calculation. Keep the worksheet in the file; at a BI claim, showing the work is worth more than the number.

Requires

Whatever exists per location: prior schedules, inspection reports, the client's descriptions, listing photos. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste what exists. Prompt: "Build the COPE profile per location from these sources: construction class and year, occupancy detail, protection (sprinklers, alarms, hydrant/station distance), exposure (neighbors, brush, flood zone if stated). Each field: value with source, or MISSING with how to get it (client, inspection, public record)."
2. Then: "Draft the per-location question list for the client, grouped so their office manager can answer most of it in one pass."

Guardrails

Construction class is never inferred from a photo caption or a year alone: it's sourced or MISSING. Protection claims (especially sprinklers) get flagged for verification; underwriters check, and wrong protection credits unwind at claim time.

Hand-back

Fill the missing fields before marketing, not after quotes arrive with assumptions baked in. A complete COPE profile is negotiating leverage: it's the difference between the underwriter's guess and your facts.

Requires

The carrier list and years needed. CLIENT FILE: insured name and policy numbers. Placeholder pattern: draft with placeholders, fill at your desk, or run it where the file stays put.

Steps

1. Prompt: "Draft a loss run request letter with placeholders: [INSURED LEGAL NAME], [POLICY NUMBERS], [YEARS], requesting currently valued loss runs for the periods listed, signed authorization from the insured attached, response requested within [X] business days. Then the variant addressed via a prior broker where we don't hold the carrier relationship."
2. Then: "Build the tracking table: carrier, sent date, chase date, received date, and the polite chase email for day ten."

Guardrails

Exact legal name and policy numbers come from documents at your desk, never memory: a mismatched name gets a letter ignored for three weeks. The letter asks for currently valued runs; undated valuations are the classic gap.

Hand-back

Five years, every line, requested on day one of the engagement: loss runs are the longest pole in every marketing timeline. Chase on schedule; carriers answer the brokers who follow up.

Requires

The loss runs, all carriers, all years. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the runs. Prompt: "Analyze: total incurred by year and line, frequency versus severity pattern, open reserves and their trajectory, the large losses with their descriptions, and the trend across the period. Flag anything odd: a spike year, reserves that grew, claims that look related. Numbers traced to run lines."
2. Then: "State what this history says plainly: improving, stable, or deteriorating, and what drives that read."

Guardrails

Open reserves are read as they stand, never discounted by optimism: an underwriter will read them at face value. Odd flags are questions, not conclusions; the story behind a spike belongs to the client.

Hand-back

Take the odd-flags list to the client before an underwriter asks: every large loss needs its story told by you, not discovered by them. This analysis feeds the loss summary (Loss summaries for underwriters) and the narrative (Submission narrative).

Requires

The loss analysis (Loss run analysis) and the client's account of the significant claims. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the analysis and the client's explanations. Prompt: "Write the loss summary for a submission: the overall pattern in two sentences, then each significant loss: what happened, what it cost, what changed because of it. Factual, unspun, and specific about the corrective actions with dates."
2. Then: "Flag any corrective-action claim that's soft ('we retrained staff'): what would make it concrete enough for an underwriter to credit?"

Guardrails

What-changed claims must be real and checkable: an invented corrective action is a misrepresentation that surfaces at the next loss. Unspun means unspun: a summary that argues gets discounted; one that informs gets credited.

Hand-back

Confirm every corrective action with the client before it ships: 'did you actually install the cameras?' is your question, not the underwriter's. The concrete version of each soft claim is usually one client call away.

Requires

The operations map (Operations walkthrough), loss analysis (Loss run analysis), and site observations. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the findings. Prompt: "Draft risk improvement recommendations: each one tied to a specific finding, stated as what to do and what it prevents, ordered by severity-prevented per effort-required. Practical for a business this size: no enterprise risk-management theater."
2. Then: "Add the tracking frame: recommended date, client response, implemented date, the columns that matter if this document is ever read in a dispute."

Guardrails

Every recommendation traces to a finding: generic best-practice padding dilutes the ones that matter. Wording is advisory and specific, never alarmist, and never guarantees that implementation prevents claims.

Hand-back

Deliver it, discuss it, and record the client's response per item: a recommendation the client declined, documented, is your E&O defense; one you never made is their lawyer's exhibit. Revisit at renewal.

Requires

An anonymized profile: class, revenue band, geography, and the current limits by line. No insured identity.

Steps

1. Prompt: "Profile: [anonymized]. For each line and limit: how businesses of this class and size typically buy, what drives buying higher (contracts, assets, judgment exposure), and whether this limit pattern looks light, typical, or heavy. Label every 'typical' claim with its confidence: pattern knowledge, not a dataset."
2. Then: "For each light flag: the plain-language conversation starter that explains the concern without predicting doom."

Guardrails

The AI's benchmarks are pattern knowledge with a date, never presented as market data: instruct it to say so in its output. Adequacy is ultimately a client decision informed by their risk tolerance; the output frames, never decides.

Hand-back

Cross-check the light flags against what you see placed in your own market: your book is the real benchmark. The conversation starters feed the recommendations letter; the client's decision gets documented either way.

Requires

The client's contracts, or the requirements extraction (Requirements extraction) already run. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the contract or extraction. Prompt: "Explain the risk transfer in this contract in plain language: what the client is agreeing to be responsible for (indemnity, hold harmless), what they're required to insure, what flows the other way, and where the contract's demands exceed what insurance typically responds to. Quote each clause explained."
2. Then: "Draft the client explanation: here's what you're taking on by signing, here's what your insurance does and doesn't cover of it, and here's the question for your lawyer."

Guardrails

The explanation describes what the wording says and where insurance responds: it never advises signing or not signing, and it names the lawyer's lane explicitly. Uninsurable obligations get flagged as exactly that, plainly.

Hand-back

The lawyer line is not a formality: contractual liability is where broker advice and legal advice meet, and the file should show you drew the line. Pair this with the compliance check (Program-vs-contract compliance check) so coverage and obligation get read together.

Requires

The completed analyses: operations map, loss read, benchmarking, recommendations. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the pieces. Prompt: "Assemble the annual risk report: their business as we understand it, their loss story, how their program compares to businesses like theirs, our recommendations with status, and what we're watching for next year. Executive summary up front, one page, written for the owner, not the industry."
2. Then: "Cut every sentence that exists to make the broker look thorough rather than to inform the client."

Guardrails

The report assembles findings already made: nothing new gets invented at assembly. The cut rule is real; padded reports teach clients not to read them.

Hand-back

Present it in a meeting; the report is the agenda, not the deliverable. This document, updated annually, is the difference between a broker and a vendor at every renewal negotiation you'll ever have with this client.

Requires

The source documents: expiring dec pages, exposure schedules, prior applications. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the source documents and the blank form's field list. Prompt: "Fill these application fields from these documents only. For every answer: the value and where it came from. Anything the documents don't answer: NEEDED FROM CLIENT, never a guess."
2. Then: "List answers that changed versus the prior application, so I can confirm each change is real and not a transcription slip."

Guardrails

Every field cites its source; an application is a signed representation, and a guessed field is a misrepresentation with your client's name on it. NEEDED FROM CLIENT is the required output for gaps: instruct the AI that inventing a plausible answer is the one unforgivable move here.

Hand-back

Review every field before it goes near the client for signature: the AI drafted from documents, but the accuracy warranty is the insured's and the diligence is yours. The changed-answers list is your ten-minute check.

Requires

The completed base application (or its data) and the supplemental's questions. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the base data and the supplemental questions. Prompt: "Answer this supplemental from the information already gathered. Three buckets: ANSWERED (with source), PARTIALLY ANSWERED (what's known, what's missing), NEEDED FROM CLIENT. Never bridge a gap with an assumption."
2. Then: "Turn the NEEDED FROM CLIENT bucket into one plain-language email to the client: every remaining question, grouped, with why it's being asked in one clause each."

Guardrails

Same law as the base application: sourced or flagged, never guessed. Supplementals often re-ask base questions in trickier wording; instruct the AI to flag any question where its answer could be read as inconsistent with the base app.

Hand-back

The consistency flags are the real E&O catch: an application and supplemental that disagree is an underwriter's favorite claim defense. One client email instead of five is the time win; send it same day.

Requires

Everything currently in the file for this placement. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the file inventory and the target forms' requirements. Prompt: "Given what these markets will ask for and what this file contains, list every missing item: what it is, which form or carrier wants it, and who has it (client, prior broker, carrier, public record)."
2. Then: "Draft the single client request: everything they owe us, grouped by how hard it is to produce, easiest first, with a one-line reason per item and a realistic deadline."

Guardrails

The list is exhaustive on the first pass: the entire point is that the client hears from you once, not five times. Reasons are one clause, honest, never scare-flavored.

Hand-back

Sanity-check the list against what you know this market actually enforces versus nominally requests: over-asking burns client patience on items nobody will read. Calendar the deadline follow-up when you send.

Requires

What's being sent (described, not pasted: "a 125, a 126, one supplemental") and how this client prefers to communicate. The completed application itself never enters a general AI.

Steps

1. Prompt: "Draft the cover email for sending [documents, described] to a client for review and signature: what each document is in one line, exactly what they need to check (the answers are drafted from their file and must be verified as true), where to sign, and the return deadline. Warm, short, zero jargon."
2. Then: "Draft the gentle chase for day three and the firmer one for day seven, both referencing the renewal clock without manufacturing panic."

Guardrails

The email must say plainly that the client is verifying and warranting the answers: convenience never dilutes whose representations these are. Placeholders for names and dates; the real details merge at your desk.

Hand-back

Attach, personalize the one line only you would write, send, calendar day three. The signed app gets your final read before it goes anywhere near a carrier.

Requires

The application data, loss summary, and what makes this account better than its class average. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the account facts. Prompt: "Write a one-page submission narrative: who they are and what they do, why this account is better than the class suggests (tenure, management, safety practice, loss trend), the honest wart and what's being done about it, and what we're asking the market for. Underwriter's reading pace: dense, factual, zero adjectives that don't carry information."
2. Then: "Cut it by a third without losing a fact."

Guardrails

Every claim in the narrative traces to the file: an embellished narrative is a misrepresentation wearing good prose. The wart stays in: a narrative with no acknowledged weakness reads as a hidden one, and underwriters price what they suspect worse than what they're told.

Hand-back

You know what this underwriter respects: adjust the emphasis, not the facts. The narrative is your signature on the account's story; read it once aloud before it ships.

Requires

The completed submission package. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the submission. Prompt: "Read this as a skeptical underwriter for [class]. List the ten questions you'd come back with: the loss that needs explaining, the exposure that looks understated, the gap in the story. For each: can the file answer it now, and how?"
2. Then: "Fold the answerable ones into a short 'anticipated questions' section for the submission. The unanswerable ones become my prep list."

Guardrails

The skeptic role is real: instruct the AI to ask the uncomfortable questions, not softballs. Folded answers cite the file like everything else; pre-answering with spin invites the question back with teeth.

Hand-back

The unanswerable list is your client call before the submission goes out, not after the underwriter asks. Ten minutes of pre-answering typically saves a week of email round-trips: that's the whole economics of this desk.

Requires

The class, the target markets, and what kinds of documents exist for this account (described, not pasted).

Steps

1. Prompt: "Build the supporting-package checklist for a [class] submission to [market types]: financials, loss runs and years, safety programs, contracts, licenses, schedules, photos, whatever this class's underwriters expect. Ordered by how much each moves the quote."
2. Then: "Draft the package cover index: numbered contents with a one-line description each, so the underwriter never hunts."

Guardrails

The checklist reflects what underwriters for this class actually use, not a maximal list that buries the good documents in filler. The index describes documents; it never characterizes their contents favorably.

Hand-back

Checking the real package against the checklist happens at your desk with the actual documents: this skill builds the standard, you enforce it. A labeled, indexed package is an underwriter courtesy that gets remembered at pricing time.

Requires

An anonymized risk profile: class, size band, geography, loss character, anything unusual. No insured name needed and none goes in.

Steps

1. Prompt: "Risk profile: [anonymized]. Reason through which market types fit: standard, specialty, wholesale-only, and why. For each suggested market type or named carrier: the appetite logic, and a confidence label (current knowledge vs general reputation)."
2. Then: "What about this profile would make a market decline on sight, and what packaging or information cures it?"

Guardrails

The AI's carrier knowledge has a date on it: every named-carrier suggestion is labeled as needing verification against current appetite guides and your own recent placements. Reputation is not appetite; instruct it to say which one it's using.

Hand-back

Your recent placement experience and appetite bulletins outrank any model's memory: the AI drafts the reasoning, your market knowledge makes the list. The decline-on-sight answer is your packaging to-do before anything ships.

Requires

The anonymized profile, plus which standard markets already passed or will.

Steps

1. Prompt: "Standard markets are out for this profile: [anonymized, with decline reasons]. Reason through the wholesale path: what specialty or E&S appetite exists for this kind of risk, what a good wholesaler adds here beyond access, and what I should require from one (markets approached disclosed, declinations documented, my client's data handled properly)."
2. Then: "Draft the wholesaler brief: the risk described for their triage, what we need, and the questions that reveal whether they actually know this class."

Guardrails

Named intermediaries carry the same dated-knowledge label as carriers. The brief describes the risk anonymized; the insured's identity travels only once you've chosen who to work with.

Hand-back

The reveal questions are the skill: a wholesaler who can't answer them is a forwarding service. Your relationships decide the pick; the AI just made sure you walk in with a standard.

Requires

Nothing from the client file: this skill builds the log structure and its language. The populated log lives in your own system.

Steps

1. Prompt: "Build a market log template for a commercial placement: market, contact, date sent, response (quoted / declined with reason / no response), follow-up dates, and a blocked-markets section. Then the standard wording for documenting a declination so it's useful at renewal and defensible later."
2. Then: "Draft the three-line note format for telling a client which markets were approached and the outcome, without editorializing on any carrier."

Guardrails

Declination language records what the market said, in their words where possible, never a characterization. The client-facing note reports outcomes; it doesn't spin a thin marketing effort into a broad one.

Hand-back

Fill it as you go, not from memory on Friday: the log's value is contemporaneity. At renewal, this document is the difference between remarketing strategically and re-annoying markets that told you no.

Requires

The narrative, the package index, and what each target market cares about. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the narrative and market list. Prompt: "Draft the submission email for each market: three short paragraphs: the account in two sentences, why it fits this market specifically, what we're asking for and when we need it. The per-market paragraph must differ for real reasons, not synonyms."
2. Then: "Subject lines: class, revenue band, effective date, in this market's preferred format if I've told you one."

Guardrails

The why-this-market paragraph is honest fit, not flattery: underwriters delete mail-merge instantly and remember it longer. Nothing in the email exceeds what the attached file supports.

Hand-back

One personal line per email in your own hand: the sentence only you could write is the one that gets it opened next time too. Log every send in the market log the same hour.

Requires

The thread with the underwriter. CLIENT FILE: their questions quote the client's specifics; strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the thread. Prompt: "Draft the reply: answer each question they asked, sourced from the file or flagged NEEDED FROM CLIENT, then one line moving us toward a quote date. Match their register: brief if they're brief."
2. For silence: "They've had it [X days], effective date is [date]. Draft the nudge: useful (one new fact or document), not needy, with a soft deadline tied to the real one."

Guardrails

Answers to underwriter questions are representations too: same sourced-or-flagged law as the application. Nudges carry something; a bare 'checking in' teaches the underwriter your emails contain nothing.

Hand-back

You know which underwriters respond to deadlines and which resent them: pick the nudge accordingly. Update the market log; a chase that isn't logged didn't happen.

Requires

Every quote received, with forms lists. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the carrier-vs-carrier comparison skill (Carrier vs carrier comparison): run it exactly on the received quotes.
2. Then: "Add a subjectivities column: every condition on every quote, and which ones are dealbreakers versus paperwork."
3. Then: "Which quote would I regret at claim time, and why, wording quoted."

Guardrails

As in Carrier vs carrier comparison: silence is never converted to exclusion, coverage before price, 'unclear' is an allowed verdict. The regret question gets a wording-based answer, not a carrier-reputation answer.

Hand-back

The regret answer is your proposal's spine: verify its quoted wording, then build the recommendation. Price enters the conversation after you know what each price buys.

Requires

The quotes. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the quotes. Prompt: "Extract every subjectivity from every quote: the condition verbatim, which quote it's on, who satisfies it (client, broker, carrier, third party), what satisfying it looks like, and the deadline stated or implied. Table."
2. Then: "Which of these, unmet at binding, void or impair coverage versus merely annoy the underwriter? Quote the wording that says so."
3. Then: "Draft the client email for their items: what's needed, why the quote requires it, by when."

Guardrails

Verbatim extraction: a paraphrased subjectivity is how one gets missed. The void-versus-annoy distinction must rest on quoted wording, because that's the distinction that ends careers when it's guessed.

Hand-back

The tracker is live until every row is closed: check it the day before binding, not the day after. An unmet subjectivity discovered at claim time is the nightmare this desk exists to prevent.

Requires

The situation, anonymized: what was quoted, what's wrong with it, what the market context is. The negotiation itself is yours; this is the prep room.

Steps

1. Prompt: "Quote came back with [terms, anonymized]. I want [target]. Build the negotiation frame: my strongest three arguments from the account's facts, the underwriter's likely counters, my fallback positions in order, and the point where walking to another market beats accepting."
2. Then: "Play the underwriter. Push back on my arguments one at a time and grade my responses."

Guardrails

Arguments come from the account's genuine strengths, never invented leverage or bluffed competing quotes: a caught bluff costs every future negotiation with that desk. The walk-away point is stated as a condition, not bravado.

Hand-back

The call or email is yours, in your voice, with your relationship capital on the table. What the rehearsal buys is the same thing it bought in BOR strategy: no surprise in the room.

Requires

The quote comparison (Quote collection and comparison) and your recommendation. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the comparison and your pick. Prompt: "Build the client proposal: their situation in one paragraph, the options compared in plain language (coverage differences first, premium after), the recommendation with its reasoning stated as trade-offs, and what happens next. A business owner reads this in five minutes and understands what they're buying."
2. Then: "Add the disclosure section: what was quoted and declined, and the line that final coverage is governed by the policy as issued."

Guardrails

The recommendation's reasoning is the comparison's findings, restated plainly: nothing new enters at proposal stage. Premium never leads a page. No superlatives about carriers or about your own firm.

Hand-back

The recommendation is your professional advice with your name on it: read the trade-offs paragraph as if the client's lawyer will someday read it to you. Then present it in person or by voice; the document supports the conversation, not the reverse.

Requires

The proposal and the quotes behind it. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the proposal. Prompt: "Build my presentation brief: the five numbers I must know without looking (limits, deductibles, premium, and the two figures this client will fixate on), the three questions this client will ask and their answers, and the one part of this program I should proactively explain before they ask."
2. Then: "Quiz me on all of it, hardest first."

Guardrails

The predicted questions come from the program's actual soft spots, not generic client behavior. Answers in the brief cite the quote wording; a confident wrong answer in the room is worse than a looked-up one.

Hand-back

The quiz is the skill: do it in the car if you have to. Walking in knowing it cold is the difference between presenting a program and reading one.

Requires

The accepted quote and the client's written instruction to bind. CLIENT FILE: this is the placeholder pattern: draft with placeholders in a general AI, fill real details at your desk, or run it where the file stays put.

Steps

1. Prompt: "Draft a bind order with placeholders: [INSURED], [QUOTE NUMBER], [EFFECTIVE DATE/TIME], confirming instruction to bind per quote [X] dated [date], listing: coverage parts bound, any subjectivities being satisfied at or after binding with their deadlines, billing arrangement, and a request for written binder confirmation by [time]."
2. Then: "List what I confirm before sending: client's bind instruction in writing, every subjectivity's status, quote hasn't expired, effective time zone."

Guardrails

Nothing binds on a verbal: the checklist's first item is the client's written instruction, and the AI must not draft around its absence. The order references the quote precisely; 'as discussed' is banned wording in a bind order.

Hand-back

Fill the placeholders from the quote document, not memory. Send, get the binder back, and run binder review (Binder review) the day it arrives: the gap between bound-as-instructed and bound-as-confirmed is where this desk's claims live.

Requires

The binder and the accepted quote. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Compare this binder against this quote: named insured, effective dates, coverage parts, limits, deductibles, forms listed, subjectivities carried forward, premium. Table: item, quote, binder, match or mismatch, wording quoted for every mismatch."
2. Then: "Which mismatches must be corrected before the policy issues versus noted for policy checking?"

Guardrails

Same law as policy checking (Policy checking): cited or NOT FOUND, no inference from form numbers alone. A binder that adds a subjectivity the quote didn't have is a mismatch, not fine print.

Hand-back

Email the carrier about must-correct items the same day, in writing: binder errors fixed before issuance are emails; fixed after are endorsements; discovered at claims are lawsuits. This review feeds straight into policy checking when the policy lands.

Requires

The binder (or quote you bound) and the issued policy. CLIENT FILE: strip named-insured details before using a general AI, or run it where the file stays put.

Steps

1. Paste both documents. Prompt: "Compare this issued policy against this binder. Build a table: item, binder says, policy says, match or mismatch. Cover named insured, dates, limits, sublimits, deductibles, forms and edition dates, endorsements listed vs attached, premium. Quote the exact wording and location for every row."
2. Then: "List anything present in one document and absent from the other."
3. Then: "For each mismatch, state which document is more favorable to the insured and why, citing the wording."

Guardrails

Tell the AI: cite the exact clause for every claim; write NOT FOUND rather than infer; never assume a standard form's content from its number alone, only from attached text.

Hand-back

Open the policy at every flagged mismatch and confirm it with your own eyes before you email the carrier. Spot-check two rows the AI marked as matching. The carrier correction request is yours to write and send.

Requires

The full issued policy. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the policy. Prompt: "Build a summary of insurance from this policy only: carrier, policy period, each coverage part with limits, sublimits, deductibles, key extensions, and the exclusions a business owner would actually care about. Plain language, one page. Cite the form and page for every figure."
2. Then: "List everything you left out that a careful broker might include, so I can decide."
3. Ask for the format you send clients (table or letter) and your standard disclaimer line.

Guardrails

Every number must come from the document, cited. No paraphrasing coverage grants into broader language than the form uses. The summary must state it is a summary and the policy governs.

Hand-back

Check every limit against the declarations yourself: a summary with one wrong number is worse than no summary. The judgment about what to highlight for this client is yours.

Requires

The policy and the client's actual question. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the policy. Prompt: "Answer this question from this policy only: [question]. Quote the exact clause(s) your answer rests on, with form number and page. If the policy doesn't answer it, say so plainly."
2. Then: "What in this policy could cut against that answer: exclusions, conditions, definitions? Quote each."
3. Then: "Rewrite the answer for the client in plain language, keeping the citation."

Guardrails

The AI answers from the document, never from what policies 'usually' say. Definitions section wins over plain-English readings: instruct it to check defined terms in every quoted clause. No coverage opinions stated as certainties; the answer describes what the wording says.

Hand-back

Read the quoted clauses in the actual policy before you hit send. If the question smells like a claim scenario, the answer goes through your E&O brain, not just the document: what you tell the client is advice, and it's yours.

Requires

The endorsement(s), plus the coverage part they modify. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the endorsement and the base form section it amends. Prompt: "For each endorsement: what it adds, what it removes, what it redefines. Quote the operative wording. Then one sentence a client would understand: after this endorsement, you are covered / not covered for X."
2. Then: "Rank these endorsements by how much they change the client's real-world protection, worst first."

Guardrails

An endorsement is read against the form it modifies, never alone: if the base wording isn't provided, the AI must ask for it rather than assume. Broadening and restricting language get equal attention; instruct it to flag every 'notwithstanding' and 'does not apply'.

Hand-back

The worst-first ranking is a draft of your advice, not the advice. Confirm the one or two endorsements that matter against the full policy, then decide what the client needs to hear and how.

Requires

The expiring policy (or its dec pages and forms list) and the renewal quote or policy. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Compare renewal against expiring. Table: item, expiring, renewal, changed or unchanged, and for every change: better, worse, or unclear for the insured, with the wording quoted. Cover limits, deductibles, forms and edition dates, endorsements added or dropped, exclusions added, premium, and any change in the basis of coverage."
2. Then: "List forms that changed edition date, and what changed inside them if both editions are provided. If not provided, list them as UNVERIFIED EDITION CHANGES."
3. Then: "Write the five changes I must tell the client about, in plain language."

Guardrails

Edition-date changes are never assumed neutral. 'Unclear' is an allowed and honest verdict. Nothing gets summarized as 'substantially similar': similar is a finding for a broker to make, not an AI.

Hand-back

Verify the worse-for-insured items in the documents, then make the remarket-or-accept call yourself. The client conversation about what changed is the job; this is its preparation.

Requires

Two or more quotes with their forms lists. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste all quotes. Prompt: "Build a comparison table across these quotes: carrier, premium, limits, deductibles, coverage parts included, notable endorsements, forms basis, subjectivities. One row per item, quotes cited. Where a quote is silent on an item another quote addresses, write SILENT, not 'not covered'."
2. Then: "List the differences that would matter in a real claim, not just on paper, and say why."
3. Then: "Draft the one-paragraph recommendation summary I'd put in a proposal, stating the trade-off plainly."

Guardrails

Silence is not exclusion: the AI never converts a missing mention into a coverage conclusion. Premium never leads the analysis; instruct it to present coverage differences first, price last.

Hand-back

The recommendation is yours: the AI drafted a trade-off summary, you decide what this client should buy and carry the reasoning. Confirm any claim-scenario assertions against the actual forms.

Requires

Current and prior policies for every claims-made line (E&O, D&O, EPL, cyber, pollution). CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the claims-made policies, both terms. Prompt: "For each line: occurrence or claims-made? If claims-made: retroactive date this term vs last term, any change in the basis of coverage, extended reporting provisions, and whether prior acts protection moved. Quote every date and clause."
2. Then: "State plainly: did the client lose any prior-acts coverage at this renewal? If any input needed to answer is missing, name it."

Guardrails

A moved retro date is an alarm, never a footnote: instruct the AI to lead with it. NOT DETERMINABLE beats a guess: this is the one comparison where a wrong 'all fine' costs a career.

Hand-back

Confirm every retro date against the dec pages with your own eyes. If anything moved, the client gets a documented conversation and you get a documented file: this seat exists for the day someone asks what you knew and when.

Requires

The client's contract (lease, MSA, construction agreement, vendor agreement). CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the full contract. Prompt: "Extract every insurance-related obligation in this contract: required coverages and limits, additional insured wording, waiver of subrogation, primary and non-contributory, notice of cancellation, certificate requirements, indemnification and hold harmless clauses, and any obligation buried outside the insurance section. Cite section and page for each. Then list sections you could not read or that reference external documents."
2. Then: "Which of these obligations are unusual or more onerous than typical for this contract type? Quote why."

Guardrails

The insurance section is never assumed to be the whole answer: obligations hide in indemnity, default, and exhibit language, and the AI is told to sweep the entire document. Anything illegible or referenced-but-absent is listed, not skipped.

Hand-back

Read the cited sections yourself for anything you'll act on, especially indemnity wording: extraction is complete when you can hand a requirements list to your team knowing where every line came from.

Requires

The extracted requirements (or the contract) and the client's current policies or program summary. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Check this insurance program against these contract requirements. Table: requirement, what the program provides, MEETS / FALLS SHORT / CANNOT VERIFY, citation on both sides. Limits and endorsement wording both count: a required endorsement the program lacks is a shortfall even if limits are fine."
2. Then: "For every shortfall: what specifically would fix it (limit increase, endorsement, coverage addition), stated neutrally."
3. Then: "Draft the client note: here's what your contract requires, here's where you stand, here's what closing the gap involves."

Guardrails

CANNOT VERIFY is mandatory wherever the program documents don't show the answer: compliance is never inferred from a certificate alone. Fix suggestions name the mechanism, not the market or the price.

Hand-back

The shortfall list is E&O-grade advice: verify each one in the documents, then deliver it in writing and keep the client's decision on file. Whether to push the client to close a gap is judgment, and it's yours.

Requires

The newly signed contract and the current program summary. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Run the extraction skill (Requirements extraction) on the new contract.
2. Run the compliance check (Program-vs-contract compliance check) against the current program.
3. Then: "Compare these requirements against the requirements already on file for this client: what is NEW or MORE DEMANDING than anything the program was built to satisfy?"
4. Then: "Draft the short client note: your new contract asks for X, your program currently does Y, here's the gap and the fix."

Guardrails

Same rules as Requirements extraction and Program-vs-contract compliance check. The comparison against existing obligations only works if you feed it the prior requirements: the AI must say so rather than pretend memory it doesn't have.

Hand-back

File the check with a date on it: the entire value of mid-term maintenance is being able to show the gap was caught when the contract landed, not discovered at a claim. The endorsement order, if needed, is yours to place.

Requires

The client's policies (or program summary) and a plain description of their operations. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the program and describe the business. Prompt: "Given this operation, list the exposures a careful broker would expect, then map each against this program: COVERED (cite where), PARTIALLY COVERED (cite the limitation), or NO COVERAGE FOUND. Do not invent exposures that don't fit the operation."
2. Then: "Rank the NO COVERAGE FOUND and PARTIALLY COVERED items by realistic severity for this business, and say what closing each typically involves, mechanism only."
3. Then: "List the questions you'd need answered about the operation to complete this analysis."

Guardrails

NO COVERAGE FOUND means found nothing in these documents, and the AI must phrase it that way: absence of evidence in a partial file is not a verdict on the program. The exposure list is grounded in the described operation, not a generic checklist dump.

Hand-back

The step-3 questions are your discovery call. Verify the top three gaps in the documents, then decide which ones this client hears about and in what order: gap analysis is a trust move, and the delivery is the broker's craft.

Requires

The full policy with forms attached. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the policy. Prompt: "List every sublimit in this policy: coverage, sublimit amount vs the main limit, where it lives. Then every exclusion, grouped: standard for this form, and unusual or added by endorsement. Quote each. Flag any exclusion added by endorsement that removes something the base form gave."
2. Then: "Which five of these would most surprise this policyholder at claim time? Explain each in one client-plain sentence."

Guardrails

Endorsement-added exclusions outrank standard ones and are flagged as such. 'Standard' is a description, never a reassurance: the AI does not editorialize that a standard exclusion is fine.

Hand-back

The five-surprises list is your client conversation. Confirm the quoted wording for anything you'll put in writing, and note in the file that limitations were reviewed and communicated: that sentence is what E&O defense looks like.

Requires

The completed gap analysis (Coverage gap identification) and your prior correspondence style, one example letter if you have it. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the gap findings and your example letter. Prompt: "Draft a coverage recommendations letter in my voice: what we reviewed, what we found, each recommendation with its reason in plain language, and a clear line asking the client to confirm which recommendations to pursue or decline. No scare tactics, no hedging into meaninglessness."
2. Then: "Add a closing paragraph that documents: these recommendations were presented on [date], and coverage decisions rest with the client."

Guardrails

Every recommendation in the letter must trace to a finding in the analysis: the AI adds nothing new at the letter stage. Tone is advisory, never alarmist: the letter should read like it was written to be re-read three years later, because it might be.

Hand-back

You sign it, so you own every line: read it against the analysis once. Send it, calendar the follow-up, and file the client's response with it: an unanswered recommendations letter is half a defense; an answered one is a whole one.

Requires

The request (who's asking, for what contract or job) and the policies behind the cert. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the request and policy summary. Prompt: "Map this certificate request against the policies: what the requester asks for versus what the policies actually provide. Table: requested item, policy supports it (cite) / does not / requires an endorsement that isn't there. Flag any requested wording (additional insured, primary and non-contributory, waiver) the program can't truthfully certify."
2. Then: "Draft the reply for any can't-certify item: what we can issue today, what the gap is, and what closing it involves."

Guardrails

A certificate never says more than the policy does: the cannot-certify flags are the entire E&O point of this skill. 'The holder insisted' is not an input the AI may honor.

Hand-back

Issue the accurate cert in your system; the mapping table is your pre-flight. The can't-certify reply goes to your client, not the holder: the client decides whether to buy the endorsement or push back on the contract.

Requires

The holder list with each cert's special wording. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the holder list. Prompt: "Structure this into a renewal-ready holder register: holder, what their cert certifies (limits shown, AI status, special wording), the contract behind it if known, and last issued date. Flag duplicates, expired projects, and holders whose wording looks nonstandard."
2. At renewal: "The program renewed with these changes: [paste changes]. Which holders' certs are affected by which change, and which reissues need more than a date update?"

Guardrails

Nonstandard wording flags are reviewed against the contract review skill (Requirements extraction), never reissued on autopilot: renewal is when last year's wrong cert becomes this year's wrong cert. Expired-project holders get confirmed dead before deletion, not assumed.

Hand-back

The affected-by-change list is your renewal-week checklist: reissue in your system, holder by holder. A clean register turns cert season from a week of archaeology into an afternoon.

Requires

The certificate (or requested wording) and the contract driving it. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Compare this certificate's wording against this contract's insurance requirements: every required element, present on the cert or missing, quoted both sides. Then the reverse: anything the cert certifies that the contract doesn't require and the policies might not support."
2. Then: "Plain-language verdict: does this cert satisfy this contract, and what specifically would a careful holder reject?"

Guardrails

Both directions get checked: missing wording fails the contract, and excess wording fails the policy. The verdict is about the documents; whether to satisfy an unreasonable contract demand is the client's business decision, framed as such.

Hand-back

This skill pairs with COI issuance (COI issuance) on every contract-driven request: run it before the holder's compliance team does. What a careful holder would reject is your client conversation today, not their stop-work order next month.

Requires

The client's ask, as they wrote or said it, and the affected policy's summary. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the ask. Prompt: "Turn this into a carrier endorsement request: policy, effective date of change, the change stated precisely (vehicle in/out with VIN placeholder, location added, AI added with the exact required wording), and what the client said versus what I'm requesting, so mismatches surface."
2. Then: "What does this change plausibly do to coverage or premium that the client should hear before we send? One plain sentence per effect."

Guardrails

The said-versus-requested comparison is mandatory: clients ask in shorthand and endorsements execute literally, and the gap between those is this desk's classic error. Premium-effect statements are 'plausibly' and 'confirm with carrier', never predictions.

Hand-back

Confirm the precise change back to the client before sending: one email now beats an endorsement reversal later. Calendar the confirmation check (Endorsement confirmation checking) for when the carrier's paper comes back.

Requires

The endorsement as issued and the original request. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Compare this issued endorsement against this request: change requested versus change made, effective dates, premium adjustment if shown. MATCH / MISMATCH per element, wording quoted. Then: does this endorsement do anything beyond what was requested?"
2. Then: "Draft the carrier correction email for any mismatch: what was requested, what was issued, what needs fixing, referencing both documents."

Guardrails

The beyond-the-request check is the sleeper: endorsements sometimes carry side effects (a form swap, a condition) nobody asked for, and those get flagged, not skimmed past. Effective dates get compared to the day.

Hand-back

File the matched endorsement against the request so the pair tells the story; send corrections the day you find them. An unconfirmed endorsement is a change that might not have happened: this ten-minute check is what 'processed' means.

Requires

The fleet change or renewal trigger and the vehicle schedule. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the schedule and the trigger. Prompt: "Build the ID card checklist for this event: which vehicles need new cards (added, replaced, renewed), the data each card needs verified (year/make/VIN against the schedule), and who at the client gets them. Flag schedule rows with incomplete VINs or mismatched descriptions before cards get cut."
2. Then: "Draft the delivery note to the client: cards attached, what changed, and the one-line ask to confirm the vehicle list is current."

Guardrails

Cards are cut from the verified schedule, never from the client's email shorthand: a card with last year's VIN fails a traffic stop, which is exactly when cards matter. The confirm-the-list ask rides along every delivery; it's free exposure-schedule hygiene.

Hand-back

Generate the cards in your system from the checked list. The flagged incomplete rows go back to the client today: this tiny seat is where fleet schedule drift gets caught cheap.

Requires

The client's question and the relevant file piece. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the question and file excerpt. Prompt: "Triage: is this answerable from the file (draft the answer, cite the source), a policy-interpretation question (route to the Policy Q&A skill), a change request (route to endorsement request), or a judgment call for the broker? For answerable ones: draft the reply in a service voice, plain, source noted."
2. Batch mode, end of day: paste the day's questions. "Triage all of these into the four buckets and draft what's draftable."

Guardrails

Interpretation questions never get answered in triage: routing them to the full Policy Q&A discipline (Policy Q&A), with its clause citations, is the skill working, not the skill failing. Drafted replies claim only what the cited source shows.

Hand-back

Review each draft against its source before sending; batch the day's judgment calls into one focused half hour instead of eleven interruptions. The triage habit is what makes the inbox a queue instead of a siege.

Requires

Who's asking and why, and the policies. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the coverage summary skill (Coverage summaries) with a recipient: run it, then add: "Tailor for a [lender/landlord/lawyer] reviewing it for [purpose]: lead with what they check first (their required coverages and limits), and include only what serves the purpose."
2. Then: "Add the standard footer: summary for reference, coverage governed by the policies as issued, certificate available on request."

Guardrails

Tailoring means ordering and selection, never enhancement: the recipient version claims exactly what the full summary claims. If the requester needs wording the program doesn't support, that's the cannot-certify conversation (COI issuance), not a summary problem to write around.

Hand-back

Same-day is the service promise this skill exists for: verify the figures against the dec pages and send. Log who received what; summaries in the wild are representations with your name on them.

Requires

The numbers only: premium amount, down payment options, rate quotes from finance companies. No client name, no account detail into a general AI.

Steps

1. Prompt: "Compare these premium finance quotes: [numbers only]. Table: down payment, installments, APR (compute it if only the finance charge is given), total cost versus paying in full, and the cancellation mechanics in plain language. Which terms would surprise a client later?"
2. Then: "Draft the client explanation: what premium financing is, what these options cost, and the one thing people miss (the finance company can cancel the policy for a missed installment). Neutral: financing is their call."

Guardrails

APR gets computed and shown even when the quote buries it: cost comparison is the whole service. The cancellation risk is stated plainly in every explanation; it's the term that hurts people. Neutrality is real: the explanation never leans toward financing.

Hand-back

The terms conversation is yours, and it's marked yours on the map: this prep just means the client hears real numbers. The signed agreement and its dates go in the file; premium finance tracking is its own back-office seat.

Requires

The client's question and the billing documents. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the question and statements. Prompt: "Reconcile: walk the charges on these documents against the policy changes that explain them (renewal, endorsement, audit, fee). Table: charge, source event, document reference. Anything unexplainable by the documents: FLAGGED for the carrier."
2. Then: "Draft the client explanation in plain language: what each charge is, why the number moved, and what we're checking with the carrier on the flagged items."

Guardrails

Unexplainable charges are flagged, never rationalized: 'probably the audit' is how billing errors get paid. The client explanation admits what's being verified; premature certainty here costs credibility when the carrier answers differently.

Hand-back

Send the carrier query on flagged items the same day as the client explanation, and close the loop when it answers. A broker who untangles billing is remembered at renewal in a way no proposal buys.

Requires

The policy's audit basis (payroll, sales, units) and the client's situation. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the policy's audit provisions. Prompt: "Build the audit prep checklist for a policy audited on [basis]: the records the auditor will request, how each should be organized (payroll by class code, overtime shown separately, subcontractor certificates gathered), and the classifications where misallocation costs real money."
2. Then: "Draft the client prep email: what's coming, what to gather, why clean records cut the bill, and the offer to review the package before the auditor sees it."

Guardrails

Prep means organized and accurate, never coached: the checklist maximizes correct classification, not minimized payroll. Subcontractor certificates get their own line; uninsured subs becoming payroll is the audit surprise that ends relationships.

Hand-back

Offer the pre-review and mean it: thirty minutes with their package before the audit catches the misallocations while they're free to fix. Calendar this at policy inception, not when the audit notice lands.

Requires

The audit statement and worksheets, plus the client's actual records. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the audit and the records. Prompt: "Check this audit: exposures used versus the client's records, classifications applied versus operations described, rates versus policy, arithmetic throughout. Table: item, auditor's figure, supported figure, MATCH / DISPUTE with the evidence for every dispute."
2. Then: "Draft the dispute letter: item by item, the auditor's figure, the correct figure, the supporting record named, requested revision. Professional, factual, no indignation."

Guardrails

Disputes rest on records, not on the bill being unwelcome: the AI disputes what evidence contradicts and accepts what evidence supports, and says which is which. Classification disputes cite the operations, not the cost difference.

Hand-back

File the dispute inside the carrier's deadline; audit rights expire quietly. Whether to fight a small difference is relationship math only you can do. Wins here are the most memorable dollars you'll ever save a client.

Requires

Client company names and renewal dates only: no policy numbers, no premiums, no personal details into a general AI.

Steps

1. Paste names and dates. Prompt: "Build my renewal calendar: for each account, the 120-day strategy date, 90-day client-contact date, 60-day submission date, 30-day decision date, walked back from renewal. Then the next 90 days as a week-by-week worklist. Flag pile-up weeks where too many accounts hit the same stage."
2. Monthly: "Updated list: [paste]. What enters the 120-day window in the next 60 days, and where are the pile-ups?"

Guardrails

Names and dates carry this seat; anything richer stays in your system. Pile-up flags are the strategic output: a calendar that doesn't warn about November is a list, not a calendar.

Hand-back

The monthly ten-minute refresh is the discipline; pile-up weeks get accounts pulled earlier, decided now. An account that surprises you at 45 days was lost at 120.

Requires

The client's class and which schedules they have (described, not pasted). Their completed answers are client data and go into your file, not back into a general AI.

Steps

1. Prompt: "Draft the annual what-changed request for a [class] client with [schedules described]: grouped questions covering operations changes, revenue and payroll movement, property and vehicle changes, new locations or ventures, new contracts signed. Answerable by their office manager in twenty minutes, with a because-clause per group so it doesn't read as homework."
2. Then: "Draft the 10-day nudge, and the shorter phone-script version for clients who won't do email forms."

Guardrails

Twenty minutes is a design constraint, not a hope: every question earns its place or gets cut. The because-clauses are honest (right values, right coverage, no audit surprises), never fear-flavored.

Hand-back

Send at 120 days per the calendar; the answers flow into the exposure schedule skill (Exposure schedules) at your desk. New contracts mentioned in the answers trigger contract review (Mid-term compliance maintenance) immediately, not at renewal.

Requires

The account's file: renewal terms offered, loss picture, service history, what changed this year. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the account picture. Prompt: "Frame the remarket-or-renew decision: what the incumbent's terms look like against what changed this year, the arguments for staying (stability, relationship, claim history with this carrier), the arguments for going to market (terms drift, appetite shifts, the changes the incumbent hasn't priced), and what remarketing costs in time and market goodwill if done idly."
2. Then: "What would each path look like a year from now if it goes badly? That's the risk I'm actually choosing between."

Guardrails

The frame presents both paths honestly: remarketing is not the default virtue, and idle remarketing spends markets. The gone-badly question gets concrete answers, not hedges.

Hand-back

The call is yours and it's the most consequential judgment on this desk: the frame just makes sure momentum didn't make it for you. Whichever path, the reasoning goes in the file in two sentences; that note is gold at next year's decision.

Requires

The expiring application, the what-changed answers, and current schedules. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the ACORD skill (ACORD applications) with a prior year: paste expiring app and the changes. Prompt: "Update this application: carry forward what the change answers confirm as unchanged, apply every stated change with its source, and flag anything the changes imply but don't state (revenue up 30% usually moves payroll: FLAGGED, not assumed)."
2. Then: "List every changed answer for my review, and every carried-forward answer that's now more than two renewals old and deserves fresh confirmation."

Guardrails

Carry-forward requires confirmation, not silence: an unchanged answer is only unchanged because the client said so this year. Implied-but-unstated changes are flagged questions, never auto-applied.

Hand-back

The stale-answers list is the quiet E&O catch: three-year-old carried-forward answers are how applications drift from reality. Same signature discipline as any application: the client verifies, you review, then it ships.

Requires

The updated application, this year's story, and the remarket decision already made. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Run the narrative skill (Submission narrative) with the renewal angle: paste the account and add: "Frame as a renewal remarket: what improved this year, why the account is at market (terms, not shopping habit), and what would win it."
2. Then run submission emails (Submission emails) for the target markets, plus: "Draft the incumbent notice that keeps the relationship: we're testing the market on terms, here's what would keep it with you."

Guardrails

The at-market reason stated to new markets must be true: underwriters price shopping-habit accounts accordingly and remember. The incumbent note is honest pressure, not a bluff; it only works if you'd actually move.

Hand-back

The incumbent conversation is timing-sensitive and yours: some carriers sharpen at a market test, some sulk, and you know which is which. Log everything in the market log; renewal remarkets write next year's leverage.

Requires

The expiring policy and the renewal terms. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the expiring-vs-renewal skill (Expiring vs renewal comparison): run it exactly.
2. Then: "Draft the client renewal letter from the findings: what stayed the same in one line, each change in plain language with its practical meaning, our read on whether the renewal is fair for the year the market had, and the ask: review and confirm."

Guardrails

As in Expiring vs renewal comparison: edition dates are never neutral, 'unclear' is allowed, nothing is 'substantially similar' by assertion. The fair-for-the-market read is framed as your professional opinion, not a guarantee of best-available.

Hand-back

The letter documents that changes were disclosed before binding: send it, get the confirmation, file both. This fifteen-minute discipline, every renewal, is the strongest E&O habit on the entire map.

Requires

The year's file: what was placed, changed, caught, and answered. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. This is the client risk report skill (Client risk reports) with a stewardship spine: paste the year's activity and add: "Include the year-in-service section: what we handled (certs issued, endorsements processed, questions answered, the gap we caught), stated as facts with counts, not self-praise."
2. Then: "Close with next year: the renewal dates, the recommendations still open, and what we're watching in their market."

Guardrails

Counts are real counts from the file: an inflated service tally is a fabricated metric wearing a bow. The gap-we-caught item appears only if there was one; manufacturing a save is banned.

Hand-back

Present it in the annual meeting; the report is why the meeting happens. A client who sees the year's invisible work itemized does not take the 'we can save you 10%' call seriously.

Requires

The account's file highlights: their season, their open items, their program's soft spots. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the account highlights. Prompt: "Find the genuine mid-term touchpoint: an open recommendation to revisit, a seasonal exposure about to matter (their busy season, weather season for their class), a certificate cluster suggesting new contract activity, or a market change affecting their line. Draft the check-in around the strongest one: useful, brief, no manufactured urgency."
2. Then: "If nothing genuine exists, say so plainly instead of inventing a touchpoint."

Guardrails

The say-so-plainly instruction is real: a hollow check-in teaches the client your calls can be skipped. The touchpoint must serve them; 'staying top of mind' is your motive, never the message.

Hand-back

Call, don't just email, for the strong touchpoints: the mid-term voice call is the retention act itself. Whatever the client mentions goes in the file the same day; check-ins are where change monitoring gets its raw material.

Requires

The account file plus fresh signals: their news, new cert requests, endorsement patterns, what they mentioned on the last call. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the file summary and the fresh signals. Prompt: "Compare what the program covers against what these signals suggest the business now does. List: signal, what it implies (new location, new service line, new contract exposure, growth past the limits' assumptions), and whether the current program contemplates it. Implications are flagged as inferences to verify, never as facts."
2. Then: "For each verified-worthy item: the one-question client email that confirms or clears it."

Guardrails

Signals imply, clients confirm: the output is a question list, not a coverage conclusion. Cert request patterns are the underused signal; instruct the AI to read them (new holders in a new city means something).

Hand-back

Run this quarterly per key account when you trigger it; the one-question emails are painless for the client and priceless for the file. A venture caught mid-term is an endorsement; caught at claim time it's a coverage fight.

Requires

The account's recent behavior from the file: response patterns, service friction, payment changes, who new is asking for certs or summaries. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the behavior picture. Prompt: "Read these signals against shopping-behavior patterns: slower replies from a previously warm contact, a summary-of-insurance request with no stated purpose, new contact asking program questions, billing friction, a competitor's name appearing. Rate the account: no signal / soft signal / active risk, with the evidence for the rating and what would confirm or clear it."
2. Then: "If soft signal or active: what does re-engagement look like this week that isn't panic? Concrete, dignified, useful-first."

Guardrails

Signals are read as probabilities, not verdicts: a busy quarter looks like cooling, and the AI must present innocent explanations alongside the risk read. Re-engagement suggestions never include discounting first; price panic teaches clients to threaten leaving.

Hand-back

You know this client's baseline; the rating only means something against it. The dignified re-engagement is a stewardship touchpoint done early, which is why this desk and 7c are the same muscle.

Requires

The situation and the account's full story: tenure, claims handled, gaps caught, service record, program quality. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the account story and what's known of the competitor's pitch. Prompt: "Build the save case: what this relationship has actually delivered (facts with dates), where the program is genuinely strong, where the competitor's likely pitch has substance versus theater, and the honest answer to 'why should we stay'. Then the meeting plan: listen first, present second."
2. Then: "What would I say if their pitch is actually better on a real point? Draft the response that keeps credibility either way."

Guardrails

The save case never disparages the competitor: the client is evaluating their own judgment, and attacks read as fear. If the competitor's point is real, the credibility-keeping response concedes and fixes it; pretending otherwise loses the account and the reputation.

Hand-back

The listen-first meeting is the save: most departures are about feeling unattended, not price, and the account story is your evidence you attended. Win or lose, the post-mortem goes through win/loss review (Win/loss review).

Requires

The loss facts as known and the policy summary. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the facts. Prompt: "Structure this into a first notice of loss: what happened, when, where, who's involved, known damage or injury, witnesses, and what's been done since. Facts only, in the client's own knowledge: no cause conclusions, no fault language, no coverage opinions. Flag every fact stated as uncertain so it's reported as uncertain."
2. Then: "List what's missing that the carrier will ask for in the first call, so we gather it now."

Guardrails

No speculation survives into a notice: 'the fire probably started in' becomes 'the fire's origin is undetermined'. Fault and cause characterizations are banned wording; a notice is a permanent document that adjusters, and later lawyers, read literally.

Hand-back

Report it the same day through the carrier's channel: speed is the one variable fully in your control. Send the client the expectations letter (Client expectations letter) within the hour; the day-one experience is where claims advocacy reputations are made.

Requires

The claim type (property, liability, auto) and your carrier-process knowledge. Placeholders for the specifics; the claim's details merge at your desk.

Steps

1. Prompt: "Draft a day-one claim expectations letter for a [claim type] claim, placeholders for names and claim number: what's been reported, who will contact them and roughly when, what the adjuster will ask for, what to document and preserve meanwhile, what not to do (no recorded statements to third parties without talking to us, no repairs beyond mitigation before inspection), and how to reach me. Calm, plain, human: they're having a bad week."
2. Then: "The variant for a liability claim where someone may sue: same calm, plus the do-not-discuss-fault paragraph."

Guardrails

Process promises stay within what you control: 'the adjuster typically calls within X days' is the carrier's timeline, labeled as typical. The don't-do list is protective guidance, never legal advice, and says so.

Hand-back

Fill the placeholders and send within the hour of the FNOL: this letter is the single highest-leverage retention document you'll ever send. Calendar the adjuster follow-up (Adjuster follow-up) before you close the file for the day.

Requires

The claim thread and its current state. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the thread. Prompt: "Status this claim: what's been requested, what's been provided, what the adjuster owes us, how long each open item has sat. Then draft the follow-up: specific about the open items and their dates, professional, and asking for a date, not an update."
2. Then: "Build the cadence: given this claim's stage, when do I chase next, and at what point does polite escalate to the adjuster's supervisor? Draft that escalation now so it's calm when I need it."

Guardrails

Follow-ups ask for dates and decisions, never vent: the adjuster controls your client's money and your tone is part of the advocacy. The escalation draft is factual timeline, zero indignation; written angry, sent never.

Hand-back

Chase on the cadence, not on client anxiety spikes: predictable pressure moves claims better than sporadic heat. Every exchange lands in the file; the timeline you're building is the escalation's evidence.

Requires

The denial or dispute letter and the policy. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste both. Prompt: "Analyze this position against this policy: the clauses the carrier relies on, quoted; the clauses that cut the other way, quoted; whether the carrier's reading requires anything the wording doesn't say; and where the position is actually solid. Honest both directions."
2. Then: "Draft the response: the policy language supporting coverage, the specific reasoning we ask them to revisit, the request for the basis of their position in writing. Firm, factual, no threats we won't act on."

Guardrails

The both-directions honesty is mandatory: fighting a solid denial spends credibility the winnable dispute will need. Nothing in the response overstates the wording; the advocate who quotes accurately gets read.

Hand-back

This is the fight the map marks yours: the analysis arms you, and the judgment of push, escalate, or advise-the-client-toward-counsel is the brokerage. Complex or large disputes get the lawyer conversation early, on the record.

Requires

The open-claims picture. CLIENT FILE: identities plus loss detail; strip before a general AI, or run it where the file stays put.

Steps

1. Paste the open claims. Prompt: "Build the claims board: claim, stage, days at stage, next action and owner (us, adjuster, client), reserve if known, and a stalled flag for anything past its stage's normal age. Order by what needs motion this week."
2. Weekly: paste updates. "Same board: what moved, what stalled, which clients need a proactive status call even though nothing changed?"

Guardrails

The nothing-changed call list is the point: silence during a claim reads as abandonment, and the board must surface clients who haven't heard from you, not just claims that need work. Stalled flags run on dates, not impressions.

Hand-back

The weekly fifteen minutes is the whole discipline. Proactive status calls when nothing changed are the cheapest loyalty you will ever buy; the board just tells you who's due.

Requires

The closed claim's story and the program. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the claim outcome and program. Prompt: "Review this claim against this program: did coverage respond as the placement intended? Where did limits, deductibles, sublimits, or exclusions bite? What did the claim reveal about the operation that the program didn't contemplate? Findings with citations, no blame framing."
2. Then: "Turn the findings into renewal actions: the changes this claim argues for, each tied to what happened."

Guardrails

The review examines the program, not the placement decisions: hindsight framing poisons the file and the findings. Revealed-exposure claims cite what the claim actually showed, never generic lessons.

Hand-back

Walk the client through it once the dust settles: a broker who closes the loop after a claim is a broker who never has to explain their value again. The renewal actions go straight into the remarket-or-renew frame (Remarket vs renew decision).

Requires

Your AMS's fields (described) and where entry errors actually bite you (certs, renewals, reporting). The client data itself is entered in your AMS, never pasted here.

Steps

1. Prompt: "Build a data entry standard for a commercial brokerage AMS: naming conventions (legal name vs DBA), required fields per record type, the fields where an error propagates (addresses onto certs, dates onto renewals), and the source-document rule: every field traceable to paper, never to memory or a phone call alone."
2. Then: "Build the ten-minute monthly QC checklist: the spot-checks that catch drift, and the error log format that shows which fields keep going wrong."

Guardrails

The standard is enforceable by a busy human: a forty-rule standard is a decoration. The propagation map is the priority order; perfection everywhere is the enemy of correctness where it bites.

Hand-back

Adopt it, tape it up, run the monthly ten minutes. The error log's pattern tells you what to fix upstream: the field that's always wrong has a broken process behind it, not a careless person.

Requires

Today's client interactions, as you'd retell them. CLIENT FILE: strip identifying details before using a general AI, or run it where the file stays put.

Steps

1. Paste the interaction. Prompt: "Turn this into a file note: date, who, channel, what was discussed, what was advised (verbatim where it matters), what was decided and by whom, what's promised next with dates. Facts and stated positions only: no characterizations of the client's mood or motives."
2. End of day batch: paste the day's calls. "File notes for each, same rules, flagging any where advice was given but the client's response wasn't captured: those need a follow-up line."

Guardrails

Advice is recorded as given, decisions as the client's: the note's job is to be boring and exact. The uncaptured-response flag matters most; advice without a recorded response is half a record.

Hand-back

Notes go in the file same day: contemporaneous is the entire legal weight of the word. The habit is the E&O policy you don't pay premium on.

Requires

How correspondence currently gets saved (described honestly) and your AMS or folder system's shape. The threads themselves stay in your system.

Steps

1. Prompt: "Design a correspondence filing taxonomy for a commercial brokerage: folder or tag structure per client (placement year, claims, certificates, advice), naming convention that sorts chronologically and says what a document is without opening it, and the rule for what must be filed versus what can die in the inbox."
2. Then: "Build the findability test: five things a defense lawyer would ask for (the advice email, the declination, the renewal disclosure), and whether the taxonomy surfaces each in under a minute."

Guardrails

The must-file rule is a short list with teeth (advice, decisions, disclosures, instructions), not 'file everything': a taxonomy nobody follows files nothing. The lawyer test is the design's acceptance criteria.

Hand-back

Migrate forward only: perfecting the archive is a trap; filing correctly from today is the win. The under-a-minute standard is what you're buying; test it quarterly.

Requires

What was offered, what was advised, what the client decided. CLIENT FILE: placeholder pattern: draft with placeholders, fill at your desk, or run it where the file stays put.

Steps

1. Prompt: "Draft a coverage declination confirmation with placeholders: on [DATE] we recommended [COVERAGE] with [LIMIT] because [REASON, plain language]; the client elected not to purchase; the client understands the exposure remains theirs; signature and date lines. Respectful, factual, no punishment tone: they're allowed to decline."
2. Then: "The companion version for a limit lower than recommended, and the note-to-file version for when the client won't sign but the advice was given."

Guardrails

The reason is stated plainly enough that the client understands what they declined: a declination the client didn't comprehend protects no one. No scare inflation; the document records advice, it doesn't campaign.

Hand-back

Get the signature when you can, file the note-to-file when you can't, and never skip it because it feels awkward: this document is the difference between 'we told them' and proving it. Revisit declined items at every renewal.

Requires

Your licenses, jurisdictions, renewal dates, and CE requirements as you know them.

Steps

1. Prompt: "Build my licensing tracker: license, jurisdiction, renewal date, CE hours required vs completed, and lead-time alarms at 90 and 30 days. Then the annual CE plan: hours needed, spread across the year so December isn't a cram."
2. Then: "List what I should verify against the regulator's current rules rather than trust from memory: requirements change, and this tracker is only as good as its inputs."

Guardrails

The verify list is mandatory output: CE rules and renewal mechanics change, and the AI's knowledge of any jurisdiction's current requirements carries a date. The tracker never claims a requirement; it holds what you confirmed.

Hand-back

Confirm requirements on the regulator's site once a year; the map marks this yours because a lapsed license is a stopped business. The 90-day alarm is the one that matters: at 30 you're already choosing between bad options.

Requires

The campaign: who's being emailed, how the addresses were obtained, what the message says.

Steps

1. Prompt: "Audit this outreach plan against CASL and CAN-SPAM basics: consent basis for each recipient category (express, implied and its expiry, existing relationship), sender identification, unsubscribe mechanism, and the records I should keep of consent. Flag every gap. Label the whole output as a working checklist, not legal advice."
2. Then: "Build the pre-send checklist version: the six questions to answer before any campaign leaves."

Guardrails

The not-legal-advice label is real and stays on: compliance frameworks shift and penalties are personal; the checklist prepares the lawyer conversation, it doesn't replace it. Implied-consent expiry gets dates, not vibes.

Hand-back

Run the six questions before every send, and keep the consent records the checklist names: CASL enforcement is record-driven. When in doubt about a category, ask a compliance professional before sending, not after.

Requires

The placement types you handle that trigger filings (surplus lines, jurisdiction-specific) and where you operate.

Steps

1. Prompt: "Build the filing obligations map for a broker placing [placement types] in [jurisdictions]: which placements trigger which filings, the typical deadlines and taxes involved, and the record each filing needs behind it. Label every requirement as to-verify against the current regulator's rules: this is a map of what to check, not a legal reference."
2. Then: "Build the per-placement checklist: at binding, the filing questions answered before the file closes."

Guardrails

Every requirement carries the to-verify label: filing rules and rates change, and a stale requirement confidently stated is worse than a blank. The actual filing happens in the regulator's portal with the real placement data; nothing client-identifying enters here.

Hand-back

Verify the map against current rules once, then trust the checklist rhythm: filings are lost to forgetting, not difficulty. Late surplus lines taxes are the classic silent penalty; the at-binding checklist is where they die.

Requires

An honest description of where client data currently lives (email, downloads, drives, AMS) and who can touch it.

Steps

1. Prompt: "Audit this data reality against good practice for a brokerage holding client financial and policy information: where documents should live (one system of record), what shouldn't sit in email, access rules, retention and disposal rhythm, and what a breach-day checklist looks like. Practical for a small firm: no enterprise security theater."
2. Then: "Build the new-document rule: when a client file arrives by email, the three steps that get it into the system of record and out of the inbox."

Guardrails

Recommendations fit a firm your size and get labeled as practice guidance, not legal compliance advice: privacy obligations vary by jurisdiction and the lawyer's lane gets named. The one-system-of-record principle is non-negotiable in the output.

Hand-back

Adopt the new-document rule this week; the backlog can wait, the habit can't. This seat is the reason the restricted stamp exists on this map: your standard for client paper is the standard you should hold every tool to, including AI.

Requires

Your invoice's required contents and where the process currently stalls. Templates draft with placeholders; real invoices generate in your system.

Steps

1. Prompt: "Build the agency-bill invoicing rhythm: the invoice template contents (placeholders), the trigger rule (invoice generates the day of binding, not the end of the month), the accuracy checklist before send (premium matches the binder, fees itemized and disclosed, taxes right), and the aging view that shows what's outstanding at 15/30/45 days."
2. Then: "Draft the invoice cover email: professional, warm, payment terms plain, and where to ask questions."

Guardrails

Fees are itemized and disclosed, always: bundled fees are a compliance problem wearing a formatting choice. The premium-matches-binder check is the one that prevents the embarrassing correction.

Hand-back

Bind-day invoicing is the whole reform: receivables age from the invoice date you control. The aging view feeds receivables follow-up (Receivables follow-up); look at it weekly, not when cash feels tight.

Requires

Your aging picture (amounts and days, placeholders for names) and your carrier payment deadlines.

Steps

1. Prompt: "Draft the receivables ladder with placeholders: day 15 friendly reminder, day 30 direct note with the carrier deadline mentioned, day 45 call script plus the letter stating consequences plainly (coverage jeopardy where true, per policy terms), each escalation professional and face-saving."
2. Then: "Build the weekly worklist format: who gets which touch this week, from the aging."

Guardrails

Consequence statements are only what the policy and law actually provide: cancellation threats that aren't real are banned, and 'coverage jeopardy where true' means checked, not assumed. Every touch leaves the client a graceful way to pay today.

Hand-back

Send on the schedule, not on your frustration: the ladder works because it's boring and predictable. The client who's always at day 45 is telling you something about the account; that's a judgment conversation, not a fourth email.

Requires

Carrier statements with insured names stripped: policy reference fragments and amounts carry the math. The named version lives in your accounting.

Steps

1. Paste anonymized rows (reference, premium, rate, amount paid). Prompt: "Reconcile: expected commission per row (premium × agreed rate) versus paid. Table: matches, shortfalls, overpayments, and rows on the statement I didn't expect or expected rows that are missing. Totals by carrier."
2. Then: "Draft the carrier query for the discrepancies: professional, row-referenced, asking for correction or explanation."

Guardrails

Rates come from your agreements, entered by you, never assumed by the AI. Overpayments get queried too: keeping a carrier's error is a relationship debt with interest.

Hand-back

Re-attach names in your accounting and book the corrections. Monthly beats quarterly: discrepancies are findable at 30 days and archaeology at 120. The map marks this yours; it's your money and worth the afternoon.

Requires

Which accounts are financed (placeholders or anonymized IDs) and each agreement's key dates. The agreements themselves stay in the file.

Steps

1. Prompt: "Build the premium finance tracker: account ID, finance company, installment schedule, the notice-of-intent-to-cancel window, and the alarm rule (flag the moment a missed installment starts the cancellation clock). Then the client outreach draft for a cancellation notice: urgent without panic, the exact cure amount and deadline, and how to pay today."
2. Monthly: paste the updated tracker. "What's inside a cancellation window right now, and what's due in the next 30 days?"

Guardrails

The cancellation clock is the whole job: instruct the tracker to treat the notice window as the deadline, not the cancellation date, because cure time evaporates. Outreach states facts and dates; shame has no place in it.

Hand-back

The map's own words: the cost of missing one is the account, and possibly a coverage gap on your watch. When the alarm fires, it's a phone call, not an email: this is the one back-office seat with a same-hour clock.

Requires

The raw signals as they arrive: appetite bulletins, marketing rep emails, your own quote outcomes (anonymized: class and result, no insureds).

Steps

1. Prompt: "Start my appetite tracker: carrier, classes they're hungry for, classes they've cooled on, evidence with date and source (bulletin, rep conversation, declination pattern), and a staleness flag on anything older than six months."
2. As signals arrive: paste them. "Update the tracker: what changed, and does any recent declination pattern contradict a carrier's stated appetite?"

Guardrails

Every appetite entry carries its source and date: an undated appetite is a rumor. Stated appetite versus behavior (the declination pattern check) is the tracker's real intelligence; instruct the AI to surface contradictions, not smooth them.

Hand-back

Feed it as signals arrive, thirty seconds at a time: appetite intelligence is a compost pile, not a report. This tracker is what makes carrier appetite matching (Carrier appetite matching) yours instead of the model's stale memory.

Requires

The raw material: industry press, carrier bulletins, rate filings coverage, and your own renewal outcomes (anonymized directions, no accounts).

Steps

1. Paste the month's material. Prompt: "Build the market briefing by line: direction (hardening, softening, stable), what's driving it per the sources, capacity notes, and where the sources disagree. Every claim cited to its source; my own renewal observations labeled as my sample, not the market."
2. Then: "The client-facing paragraph version: what this means for a business renewing in the next quarter, honest, no fear framing."

Guardrails

Sources disagreeing is a finding, not a problem to average away. The client paragraph never uses market conditions to pre-excuse a bad renewal or manufacture urgency: it informs, and the sales restraint is the credibility.

Hand-back

The client paragraph feeds your LinkedIn posts, renewal letters, and stewardship reports: one briefing, four uses. Your own renewal sample is the ground truth the press doesn't have; weight it accordingly.

Requires

The risks you couldn't place this year (classes and reasons, anonymized) and the public material on candidate markets.

Steps

1. Prompt: "I keep failing to place: [classes, anonymized]. From this material on candidate markets [paste their public appetite pages, AM Best info, program descriptions], which are genuine fits: appetite match, financial strength as rated, access route (direct appointment, wholesale, program). Every fact cited; label reputation versus verified."
2. Then: "Build the diligence checklist before I send them anything: ratings verified current, appetite confirmed by a human there, claims reputation asked about, contract terms to check."

Guardrails

Financial strength claims cite the rating agency and date, never memory. The diligence checklist is mandatory output: a new market is a counterparty decision, and the AI's job is making sure enthusiasm doesn't skip the checks.

Hand-back

The confirmed-by-a-human step is yours: appetite pages are marketing until a person there says yes to your class. One new genuine market a year, properly diligenced, beats five appointments you never use.

Requires

The raw notices: bureau circulars, carrier form bulletins, industry coverage of wording changes.

Steps

1. Paste the notice or bulletin. Prompt: "Summarize this form change: what wording changed (old versus new, quoted where the source shows it), which lines and forms it touches, effective dates, and who in a commercial book it bites: which classes, which coverage expectations."
2. Quarterly: paste the accumulated summaries. "The quarter's form-watch digest: the three changes that matter most for a commercial book, and the renewal-season checks they create."

Guardrails

Old-versus-new is quoted from the source or marked as unavailable: paraphrased wording changes are how the watch itself introduces errors. Who-it-bites stays at class level; account-level implications get worked in the file, where they belong.

Hand-back

The renewal-season checks flow into your comparison discipline (Expiring vs renewal comparison, Renewal comparison): a form watch that doesn't change what you check at renewal is trivia. This is the reading that separates the technical desk from the quote fetchers.

Requires

The upcoming meeting: which carrier, which people, what your book has done with them (anonymized volume and outcomes).

Steps

1. Prompt: "Prep my carrier meeting: what my numbers with them say (volume, hit ratio, retention, anonymized), the story those numbers tell from their side of the table, what I want (appetite intel, a hard-risk conversation, better terms access), and the questions that make their marketing rep actually useful."
2. Then: "The follow-up note after: what was said, what was promised both directions, logged so next year's meeting starts from the record."

Guardrails

The their-side reading is the prep's real value: a broker who knows how the carrier sees their book negotiates differently. Promises get logged verbatim both directions; relationship capital is a ledger too.

Hand-back

The map marks this yours because relationships don't delegate: the prep just means you spend the room on the relationship instead of the numbers. The rep who returns your calls in a hard market was earned in meetings like this one.

Requires

Public signals: competitor content, their job postings, what prospects mention hearing, industry moves.

Steps

1. Paste the signals. Prompt: "Read this competitor picture: who's active in my classes, what their public pitch emphasizes, what their hiring says about their direction, and where their pitch is strong versus where it's theater. No speculation beyond the signals; gaps are listed as unknowns."
2. Then: "What does this mean for my own positioning: where I'm differentiated for real, and where I'd lose a bake-off today?"

Guardrails

Public signals only: nothing that involves misrepresentation, pretexting, or anyone's confidential material, and the AI refuses inputs that smell like it. The where-I'd-lose answer is required honesty; a competitor watch that only flatters is a mirror.

Hand-back

The lose-today list is your improvement queue, and it feeds the save-case skill (Competitive response) before you ever need it in a fight. Check the picture quarterly; obsessing weekly is how you start selling against them instead of for the client.