Skip to main content
DoneThat

AI Adoption GuideInsuranceRenew

Personalized renewal communication

LLM drafts renewal communications that explain rate changes, coverage updates, and next steps in plain language.

Insurance processQuoteUnderwriteBindIssueBillServiceRenewClaim

By Don, DoneThat’s AI coach · updated

What a sendable renewal letter contains

A quality renewal letter explains issued renewal terms in plain language and, when it compares old and new, cites the in-force schedule for what is on the books today. Quality is not tone, length, or a friendlier greeting. Quality is a letter a renewal-comms or agency-ops lead can hand to comms without inventing a number, filling a gap with a guess, or implying the letter already went out.

Rate changes, coverage updates, and next steps may all appear in the same packet. Each factual sentence needs a cite to issued terms. If a given term has not issued, that sentence stays empty. The rest of the packet can still move. Comms still sends.

This is a draft step, not a rewrite of last year's notice. Load what underwriting actually issued, draft against that plus the in-force schedule, leave blanks where the issue is incomplete, and stop before send.

Policy admin and agency systems (Guidewire, Duck Creek, Salesforce, Applied Epic, and the rest of that class) already hold the documents this page depends on. This page does not replace those systems. It sits between issued terms and the comms queue.

Pull issued terms and the in-force schedule first

Load two sources before the model writes a word: the issued renewal terms, and the in-force schedule. Issued terms are the renewal the carrier has actually produced. The in-force schedule is what the policyholder has today. A letter that explains a change without both is either silent on the old book or inventing the new one.

If issued terms are not on the file, do not draft the paragraphs that depend on them. You can still assemble a shell: greeting, policy number, and a renewal effective date only if that date is issued. Leave a blank block for the missing terms. Do not backfill from a quote, a rating worksheet, or last year's letter.

Rate change is the paragraph teams most often want to complete. If the issued terms state a new premium as a dollar amount, cite that dollar amount. If they also state a change in dollars, cite that. If they do not state a percent, do not compute one and put it in the letter. A computed percent is not an issued term. The policyholder will treat it as one.

Coverage updates follow the same rule. An endorsement, limit, deductible, or exclusion belongs in the letter only when it appears on the issued renewal. Comparing it to the in-force schedule is how you say this is new without guessing. If the endorsement is still in underwriting, that paragraph is blank. Comms can send the rest or hold the packet. The model does not fill the hole.

When the rate itself is still being set, that work is dynamic renewal pricing, not this draft. This page starts after a term is issued.

Draft with cites; empty stays empty

Write plain language, attach a cite to every factual line, and leave a line blank when there is no issued term to cite.

A cite is a pointer a reviewer can follow: the issued renewal document and the clause or schedule line, plus the matching in-force schedule line when the sentence is a comparison. A sentence with no terms cite is not a draft. It is copy. Reject it the same way you would reject an uncited coverage answer from a policy RAG coverage assistant.

Next steps (pay-by date, documents to return, how to reach the agency) are allowed only when they appear in the issued packet or in a comms template the carrier already approved. Do not invent a due date from a rating calendar. Do not invent a portal address. Do not add a coverage pitch because the household looks underinsured. A gap on the schedule is a different work item: coverage gap cross-sell at renewal.

One illustration, not a measured result. A HO-3 renewal whose issued terms show dwelling 450,000 and a wind/hail deductible of 2 percent of dwelling, effective 1 November. The in-force schedule shows dwelling 420,000 and a 1 percent wind/hail deductible. The issued terms also show a renewal premium as a dollar amount. They do not show a percent change.

A sendable draft says the dwelling limit will be 450,000 (cite issued terms), up from 420,000 on the in-force schedule (cite schedule); the wind/hail deductible will be 2 percent of dwelling (cite issued terms) compared with 1 percent on the in-force schedule (cite schedule); and the renewal premium is the issued dollar figure (cite issued terms). The rate-change percent line is blank. There is no issued percent.

If the wind/hail deductible had not yet issued, that entire deductible paragraph stays empty. The dwelling and premium paragraphs can still draft. Comms decides whether the packet goes out incomplete or waits. The model does not write "your deductible may change" as a substitute. That sentence has no terms cite.

Hand the draft to comms; do not treat it as sent

The output is a draft in the comms queue, not a sent letter. Do not auto-send. Renewal-comms and agency-ops still own channel, timing, suppression, and the legal review the carrier requires.

Treating the draft as sent is a failure even when the text is accurate. Downstream systems will show the policyholder as notified. Service and agency staff will answer as if the letter went out. If comms later holds the packet because a term is still missing, the file now shows a notification that never happened.

Queue the draft against the same policy and renewal ID the issued terms use. Salesforce or Applied Epic may be where the agency sees the activity. Guidewire or Duck Creek may be where the terms live. The integration pattern varies. The rule does not: the letter does not leave the building until comms sends.

If a churn flag is on the file, that score can change who reviews the draft. It does not change the issued premium or the cited terms. Scoring is churn propensity scoring at renewal, not a reason to soften a number in the letter.

Failure modes that look finished

Three defects pass a casual read and fail a quality review.

A line with no terms cite. Common forms: your premium increased modestly; we updated your coverage to better protect you; your rate changed because of the market. None of those sentences point at issued terms or the in-force schedule. Modest, better, and because-of-the-market are not issued. Strike the line or replace it with a cited fact. If you cannot cite, leave it empty.

Treating the draft as sent. The model, the workflow, or an eager integration marks the activity complete when the draft lands in a folder. Comms has not sent. Keep draft and sent as separate statuses. Empty paragraphs do not block a draft from existing. They do not authorize send.

Inventing a rate-change percent. Teams do this when the issued terms show two dollar premiums and someone wants a headline. The percent is easy to calculate and easy to get wrong (fees, taxes, installment versus annual, midterm endorsements). More important: it was not issued. If the carrier wanted a percent in the letter, it would be on the issued terms. Leave the percent blank.

A fourth defect sits next to these: filling an unissued coverage update with last year's wording or with a quote the underwriter has not bound. The letter then describes a policy the carrier has not issued. Empty stays empty.

What this letter is not for

This letter is not the rating decision. If premium is still moving, wait for issue. This letter is not a coverage Q&A. A policyholder who asks whether water backup still applies needs retrieval against the issued form, not a helpful paragraph the model added. This letter is not the cross-sell and not a retention script. A gap or a churn flag opens a different work item. It does not rewrite the cited terms.

The operating sequence is short. Load issued terms and the in-force schedule. Draft in plain language with cites. Leave blanks where a term is not issued. Place the draft in the comms queue. Comms sends. Stop there.

Is this worth automating for you?

Whether this pays back depends on how much time it takes your team today. Most teams estimate that from memory, and the estimate is usually wrong in one direction or the other.

DoneThat reconstructs where the time actually went, with no timers to forget, so you can measure the baseline before committing to a project and check the gain afterward.

Measure the baseline first