Most of a renewal is last year’s account with a few things changed. The hours go into finding last year’s file, chasing the updated statement of values and the currently valued loss runs, and comparing two years side by side before anyone can think about the account. An agent can do that assembly. It starts on a date or a signal, gathers the documents, extracts them into the same structure as last year, works out what moved, and drafts the renewal pack and the questions for the client. The account executive reviews, edits and approves; nothing goes to a client or a market without a person; and every source, draft and approval is recorded for the client file and for E&O. Below is the timeline, what belongs in the pack, the workflow, the guardrails, the architecture and a pilot plan for one practice.
Previous: Part 2 — From ACORD forms, SOVs and loss runs to clean data
A renewal is mostly a scavenger hunt
Ask an account executive what a renewal involves and the honest answer starts with a search. Find last year’s submission and the terms that were bound. Ask the client for the updated SOV and exposure schedule. Request currently valued loss runs from the carrier of record. Open both years’ workbooks side by side and work out what changed. Only then does the part that needs a broker begin: deciding what the changes mean for the program, what to tell the client, and which markets to approach.
That order is the problem. The judgment work is gated behind assembly work, and assembly work is exactly what slips when an account team is carrying a full book. The renewal that should have started at 120 days starts at 45, the client gets a list of questions in a hurry, and the markets get a submission that is complete on paper but late.
The observation the rest of this post builds on is simple: at renewal you already know the account. You have last year’s locations, values, loss history and terms. The valuable question is not “what does this account look like?” but “what changed since last year, and what does it mean?” If the gathering, extraction and comparison happen automatically, the account team opens the renewal already looking at the changes that matter.
This is not a new idea for underwriters. On the carrier side, InsightXtract already pulls updated SOVs, loss runs and exposure data, extracts them into the prior term’s structure and gives the underwriter a year-over-year delta view — described in Renewals in the workstation. This post applies the same pattern to the broker’s side of the renewal, where the pack is prepared rather than reviewed.
The renewal timeline, and where the hours go
Every firm runs its own calendar, and large or complex accounts start earlier. The shape below is generic, but the point holds for all of them: the work that decides the outcome happens early, and the early stages are where assembly crowds out judgment.
What belongs in a renewal pack
A renewal pack is the account team’s working document: what the client has, what changed, and what still needs an answer. The agent drafts it; the account executive owns it. These are the sections we build, with where each one comes from.
| Section | What it shows | Source |
|---|---|---|
| Exposure changes | Payroll, revenue, square footage, vehicle counts and other rating bases, this year against last | Client exposure schedule, prior submission |
| Updated SOV | Locations added or removed, value changes, construction and occupancy updates | Client SOV, extracted and normalized |
| Loss runs | Currently valued loss runs per carrier and line, with new and developing claims flagged | Carrier loss runs |
| Claims summary | Open claims, large losses, reserve movement and anything the client should be ready to explain | Loss runs, the firm’s claims advocacy notes |
| Prior terms | Limits, retentions, key endorsements, exclusions and open subjectivities from the expiring program | Bound policies and binders |
| Market feedback | What each market said last year, declinations and their reasons, appetite notes | Placement notes and correspondence |
| Coverage gaps | Coverages the client does not buy that similar companies typically carry | Peer benchmarks and insured intelligence |
| Client changes | Acquisitions, divestitures, new subsidiaries, financial-distress signals, leadership changes | Insured intelligence signals |
| Questions for the client | Gaps and inconsistencies the agent could not resolve from the documents | Generated from the comparison |
The last three rows are where insured intelligence earns its place. InsightPulse Insurance 360 fuses internal systems with public datasets and licensed third-party feeds. Two of the signals it describes are the ones a broker most wants before a renewal conversation: a financial-distress signal built from filings, credit-score movement and news, and a coverage-gap detector that compares an insured’s program with similar-sized peers. Corporate linkage that shows subsidiaries the program does not cover belongs in the same section.
Every line in the pack cites its source. A value in the exposure section links to the cell in the client’s workbook it came from; a coverage-gap note names the benchmark it was compared against. If the account executive cannot click through to the evidence, the line does not belong in the pack.
The agent workflow, step by step
The workflow has seven steps. The first five run without a person; the sixth is the account executive’s; the seventh happens every time, whatever was decided.
Trigger
Most renewals start from the expiry date in the agency management system: the agent picks up every account crossing the firm’s start threshold, for example 120 days. Some start from a signal instead — an acquisition, a new location in the client’s filings, a financial-distress alert — because a material change is a reason to start early, whatever the calendar says.
Gather
The agent collects what the pack needs: last year’s submission and bound terms from the AMS and document store, the client’s updated SOV and exposure schedule from the shared folder or portal where they land, and loss runs from the carrier correspondence. Where a document is missing, it drafts the request for the account team to send, rather than sending it itself.
Extract and validate
Each document is extracted into the same structure as last year’s: locations with construction, occupancy and values; claims with dates, causes, paid and reserved amounts; exposures by rating basis. Values are normalized before anything is compared, and low-confidence fields go to a person instead of into the pack. Part 2 covers this step in detail.
Compare year over year
With both years in one structure, the comparison is arithmetic: locations added or removed, values up or down, new and developing claims, exposure swings, terms that differ from what was bound. Changes are ranked by materiality, so the new location and the large new claim appear before a wall of unchanged rows.
Draft the pack and the client questions
The agent writes the pack sections in the table above, each line linked to its source, plus a list of questions for the client: the inconsistencies it found and the gaps it could not fill. It also notes what last year’s markets said, so the account team starts marketing with the history in front of them.
Review, edit and approve
The account executive opens a pack that is already assembled. They correct what is wrong, decide what to raise with the client, and approve the questions before anything leaves the firm. The agent never sends to a client or a market.
Record
Every source the agent read, every draft it produced, every edit and the final approval are recorded against the account. When someone asks a year later why a coverage was or was not recommended, the answer is in the file.
The architecture
The agent does not replace the systems a brokerage already runs. It sits between them, reads through governed tools, and writes its drafts and records back to the account.
- Systems of record stay where they are. The AMS and CRM hold accounts and bound terms; document stores and shared folders hold SOVs and policies; email and portals carry loss runs and market correspondence.
- InsightXtract does the documents. It reads SOVs, exposure schedules and loss runs into a stable structure with confidence scores, so year-over-year comparison is trustworthy.
- InsightPulse supplies the signals. Financial distress, corporate linkage and coverage-gap benchmarks feed the client-changes and coverage-gap sections. Company and broker data stays current as distribution consolidates — the reason we wrote Building a trusted broker hierarchy.
- The agent works through governed MCP tools. Each tool does one thing — fetch prior terms, list documents for an account, get loss runs, get signals — and enforces its own permissions outside the model.
- InsightLense keeps the audit trail. The reasoning, the authority the agent acted under and the records it touched are kept as one thread per renewal.
Guardrails
A renewal pack carries client financials, loss history and the firm’s view of the client’s exposures. The guardrails are what make it acceptable to let an agent near that.
| Guardrail | How it works |
|---|---|
| On-behalf-of permissions | The agent acts as the account team that owns the client. It cannot read an account the account executive could not open themselves. |
| No outbound without a person | The agent drafts emails, questions and requests. Only a person sends anything to a client, a carrier or a wholesaler. |
| Citations or nothing | Every value and statement in the pack links to its source document, field or signal. Uncited content is removed before review. |
| Confidence routing | Low-confidence extractions are shown as questions, never as facts. |
| Read-only tools | The agent cannot change the AMS, policies or client records. It writes only to its own drafts and the audit trail. |
| Recorded approvals | Each approval records who approved, what was changed and when, for the client file and for E&O. |
The failure modes we plan for
1. Comparing documents that were never aligned. A location written differently this year looks new, and a revalued building looks removed and added. Normalize and match by a stable key before comparing, or the delta is noise.
2. Stale loss runs presented as current. A loss run valued months ago reads like this year’s. Show the valuation date on every loss section and flag anything older than the firm’s threshold.
3. Signals treated as facts. A financial-distress signal is a reason to ask a question, not a statement to put in front of a client. The pack labels signals as signals and keeps their sources.
4. The agent that sends. Convenience pressure will eventually ask for “just send the client questions automatically.” Hold the line: drafts only, and a person sends.
5. Edits that do not feed back. If corrections live only in a downloaded spreadsheet, the same mistake comes back next year. Corrections go back to extraction.
What to measure
We agree the measures with the practice before the pilot starts and take the baseline from the current renewals, so the comparison is the firm’s own and not a vendor’s claim.
- Days to pack ready: the gap between the renewal start date and the account executive having a reviewable pack.
- Pack completeness: the share of sections populated with cited sources at first review.
- Corrections per pack: values the account executive changed, by section — the clearest signal of extraction quality.
- Client questions answered before marketing: whether open questions are closed before submissions go to markets.
- Hours per renewal: the account team’s own estimate of assembly time, before and after.
A pilot on one practice
| Weeks | Work | Exit gate |
|---|---|---|
| 1–2 | Pick one practice and line, map where renewal documents live, agree the pack template and the measures | Baseline taken on recent renewals |
| 3–4 | Connect the AMS and document stores read-only; extract last year’s documents for the pilot accounts | Prior-term records match the bound terms |
| 5–6 | Run the agent on upcoming renewals in parallel with the team’s usual process | Account executives review every pack |
| 7–8 | Switch the pilot accounts to agent-first preparation; tune extraction from corrections | Measures compared with the baseline |
Readiness checklist
| Before you start | Ready when |
|---|---|
| Scope | One practice, one line and an agreed list of pilot accounts |
| Prior terms | Bound terms for the pilot accounts are retrievable from the AMS or document store |
| Document locations | Every place SOVs, exposures and loss runs arrive is known and connectable read-only |
| Pack template | The practice has agreed the sections and their order |
| Permissions | The agent’s access mirrors the account team’s, and nothing more |
| Approvals | Who approves a pack, and what is recorded, is written down |
| Measures | Baseline taken on recent renewals before the agent runs |
The bottom line
A renewal is mostly last year’s account with a few things changed, and the hours go into finding out which things. An agent can do that finding: gather the documents, extract them into last year’s structure, work out what moved, add what the outside world knows about the client, and draft the pack with every line cited. What it cannot do, and should never be asked to do, is decide what the changes mean for the client’s program or speak to the client and the markets. That stays with the account executive, who now starts the renewal with the work already assembled and the evidence one click away.
See the full series on the AI for Insurance Brokers page.