Insurance AI agent · Renewal comparison
Renewal-Comparison & Policy-Checking AI Copilot
Compare the expiring policy with the renewal, or the issued policy with what was quoted, and report every difference with the wording behind it — a broker decides what it means.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Coverage advice
- Broker or UW
- After launch
- Agent Care
What this agent does
Reports the differences, advises on none
In
- 01
Take both documents as they stand — the expiring policy or the quote, and the renewal or the policy the carrier issued.
- 02
Confirm the expiring side is the version that was in force, mid-term endorsements included, and stamp what was read.
Reason
- 03
Match every form, edition, endorsement and schedule on each side, and mark the wordings that fit no standard form.
- 04
Set clause against clause — limits, sub-limits, aggregates, deductibles, exclusions, conditions, definitions and schedules.
- 05
State what each difference would do if a claim were made, and quote both wordings so a reviewer can read them.
Decide
- 06
Where nothing has moved, the comparison still goes to a broker — an empty report is evidence of a check, not a result.
- 07
Anything unparsed, non-standard or restructured goes to a broker with the gap named rather than quietly closed.
Out
- 08
Hand over a difference report — each change, the clause on both sides, and the consequence written in claims terms.
- 09
Retain both documents as read, the editions matched, the differences found and the advice the broker gave the client.
Product statement
The agent reports what changed and points at the wording. A qualified broker or underwriter decides what it means and advises the client — and a report with nothing on it is a statement about two documents, never a clean bill of health.
Example workflow
One renewal, end to end
- 1Renewal receivedThe expiring policy or the quote on one side, the renewal or the issued policy on the other
- 2Versions confirmedEach side resolved to a policy, period and insured, with the expiring version checked against the endorsements issued mid-term
- 3Forms and editions matchedEvery form number, edition date, endorsement and schedule paired across the two documents, and non-standard wordings marked as such
- 4Clauses comparedLimits, sub-limits, aggregates, deductibles, exclusions, conditions and definitions set against each other, both wordings kept
No human action required
Stages 1 to 4 run without a person in the loop — matching, comparison and consequence-writing are finished before a broker opens the report. Nothing has been advised.
Reaches the broker as a complete comparison.
Reaches the broker with the gap named.
Broker, account handler or underwriter
Reads the wording on both sides, decides which differences matter to this client, and advises them in their own words.
Accept · Correct · Query carrier
Checked — handed back
- 6Report handed to the brokerEach difference with the clause on both sides and what it changes at claim, and nothing marked as acceptable
- 7Outcome evaluatedDifference recall, false differences, consequence completeness, edition currency and review time, by document type and line
Reviewer corrections
Every difference a reviewer adds or strikes out is counted, in both directions.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Telling a client the cover is adequate.
- Deciding which differences matter to this client.
- Marking a difference acceptable, or closing it.
- Signing the policy check or the renewal file.
Automation boundary
Agent acts unaided
- ✓Match forms, editions and endorsements on both sides.
- ✓Set clause against clause and quote both wordings.
- ✓State what each difference would change if a claim were made, in claims terms.
- ✓Name what it could not read, and every wording that fits no standard form.
Write actions run only inside the approval boundaries agreed during implementation. Advising a client on cover, and signing the check, are not among them, on any tier, in any configuration.
- Recommending one quoted alternative over another.
- Reporting no differences without a broker reading it.
- Feeding comparison output into pricing or retention.
- Changing form libraries, edition data or comparison rules.
Example output
One renewal comparison, annotated
Every difference the agent reports is attached to the clause it came from, on both sides.
Comparison output · single renewal
Illustrative example- Comparison
- Expiring vs renewal
- Cover
- Commercial combined
- Documents read
- Both, versioned
- Differences found
- Four, all quoted
- Confidence
- 92%
- Advice given
- None — broker's
As received
The expiring policy as it stood at expiry, and the renewal wording as the carrier issued it.
Wording cited
- Form edition, both sides
- Endorsement schedule
- Sub-limit table, quoted
Why four, not one
One endorsement changed edition, and a sub-limit inside it moved. Both are reported.
What the score decides
How much of the wording a reviewer reads at source. Never that a change is acceptable.
Action
AcceptCorrectQuery carrierValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every renewal
From the renewal list, or a policy as issued
Read the clause, not the schedule
Form and edition dates on both sides, every endorsement and schedule attached to them, and the clause text itself — not the schedule page, and not last year's summary of it.
Read every page, every time
Both documents are read end to end at renewal volume, so the check does not thin out in the weeks when the most policies land at once.
Say what a difference costs
Each change is reported with what it would do if a claim were made, so a broker weighs it rather than initialling a list of clause numbers.
Build an evidence trail
Retain both documents as read, the editions matched, the differences reported, the ones a reviewer added or struck out, and who signed the check — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Broking & policy admin
- Applied Epic
- Acturis
- Guidewire
- Duck Creek
- Sapiens
Forms & edition data
- ISO forms
- LMA and market wordings
- Carrier form libraries
- edition dates
Document management
- SharePoint
- iManage
- Policy PDFs and scans
- attachments
Agent
Renewal comparison & policy checking
- Matches editions
- Compares clauses
- Reports differences
Quote, slip & client file
- Quote and binder capture
- market slips
- Client file and CRM
- e-signature
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the wording and the advice
Each control wraps the one inside it. A difference clears every layer before it reaches a broker, and the advice to the client sits outside all six.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether a difference may stand
- L3
- Keeps an adequacy view out of the agent
- L4
- Puts the advice with a qualified broker
- L5 – L6
- Keep the trail and pull automation back
How Nestack evaluates it
Evaluate the differences found — and the ones a reviewer added.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the difference report a broker opens
- E1Final-output evaluationDid it find every difference the two documents actually contain?
- E2Step-level evaluationWere forms and endorsements matched by edition, from a current library?
- E3Tool evaluationDid it read the version of the expiring policy that was in force?
- E4Report-quality evaluationDoes every difference carry its consequence, and how much is noise?
- E5Slice evaluationHow does recall change across document types and lines of business?
- E6Business outcomeWhat did reviewers add or strike out, and how long does a policy take?
Floor — a check the broker can still stand behind when it is questioned
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycle
Direction of processing →
01 · Document intake
1 mode
- RN-01
Wrong expiring version
Compared against a version a mid-term endorsement replaced.
Stage pairs
The expiring version and the renewal as issued
02 · Form matching
2 modes
- RN-02
Edition change missed
Same form number, later edition, an exclusion no one read.
- RN-03
Manuscript wording misread
A bespoke endorsement mapped to the nearest standard form.
Stage matches
Forms, editions, endorsements and schedules
03 · Clause comparison
2 modes
- RN-04
Sub-limit change unreported
The limit holds; the sub-limit inside it does not.
- RN-05
Schedule difference lost
The two documents structure the schedule differently.
Stage compares
Limits, deductibles, exclusions and conditions
04 · Report / handover
1 mode
- RN-06
Difference without consequence
Reported as a clause number, so it is initialled through.
Stage hands over
The difference report the broker works from
05 · Drift / Version
1 mode
- RN-07
False differences in volume
Noise grows with the library, and reviewers start skimming.
Stage tracks
Form libraries, editions and how reports are used
- Sev-1 · a real change reaches the client unseen
- Sev-2 · the report overstates the check
- Sev-3 · the report is noisy, review absorbs it
Affected slices
The weakest checks land on the smallest accounts
Scanned wordings, bespoke endorsements and restructured schedules cluster in small commercial and non-standard business, so the clients least able to absorb an uncovered loss get the least reliable check. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Scanned and faxed policies | 6.9% | 3.5× | Review | |
| Manuscript and non-standard wordings | 5.0% | 2.5× | Review | |
| Schedules restructured by the carrier | 3.7% | 1.9× | Watch | |
| Repeat renewals, same form set | 1.6% | 0.8× | Normal |
Bar: missed-difference lift vs. repeat-renewal baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Coverage gaps surface at claim, years after the check
The question is always the same: did the check miss the difference, or did it name one to a reviewer who had stopped reading?
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Difference recall or the false-difference rate moves in one document cohort.
- 02
Diagnose
Traced to the version pinned, the edition matched, the comparison or the wording of the report.
- 03
Improve
The form library, matching rule or report format is changed, re-approved by the broking lead and version-linked.
- 04
Verify
Re-run against held-out renewals, including every difference a reviewer had to add.
- 05
Learn
The missed difference becomes a regression case, with the clause from both sides kept.
Learn → Detect
The return edge. Every cycle asks where the report went — a comparison used to pick which accounts are kept or priced is an underwriting input, proxy variables and all.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, document and form sourcing, comparison and consequence work, evaluation, review workflow, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Document sources and version records.Week 1, Week 2
- 03Broking, policy-admin and document APIs.Week 1, Week 2
- 04Form library, edition data and currency checks.Week 2, Week 3
- 05Non-standard and manuscript wordings.Week 2, Week 3
- 06Clause comparison and schedule alignment.Week 3
- 07Consequence annotation in claims language.Week 3, Week 4
- 08The no-advice constraint and output use limits.Week 4
- 09Broker review workflow and report-format sign-off.Week 4, Week 5
- 10Recall, false-difference and cohort evaluation.Week 4, Week 5
- 11Quote, slip and e-signature handover.Week 5, Week 6
- 12Observability, deployment and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne line, one form set | ProductionProduction broking integration | AdvancedMulti-line / multi-market |
|---|---|---|---|
| Introduced at Pilot | |||
| Expiring-to-renewal comparison | ✓ | ✓ | ✓ |
| Form and edition matching | ✓ | ✓ | ✓ |
| Clause-level difference report | ✓ | ✓ | ✓ |
| Consequence annotation | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Quote-to-policy checking | — | ✓ | ✓ |
| Non-standard and manuscript wordings | — | ✓ | ✓ |
| False-difference control and noise budget | — | ✓ | ✓ |
| Sign-off record and audit trail | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Renewal-season throughput | — | — | ✓ |
| Multi-market and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the lines and markets in scope, the form libraries and edition data available, broking-platform and document-store integrations, renewal volume, review workflow controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01The lines, markets and renewal types in scope
→Workflow discovery and automation-boundary definition
Week 102Where the expiring policy and its endorsements sit
→Document sources and the expiring-version record
Week 103Access to the broking platform and document store
→Broking, policy-admin and document-store integration
Week 204Your form library and the editions you hold
→Form and edition matching, and non-standard wording handling
Week 205Your written rule on what a check must report
→Consequence annotation and the no-advice constraint
Week 306Real renewals, including differences you once missed
→Evaluation suite, regression cases and failure-mode testing
Week 407Named brokers and an account-handling lead
→Review workflow, report format and the sign-off record
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. The overlap in week 5 is where live renewals are checked twice, by the agent and by your own people.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Lines and renewal types in scope, and who signs a check
W2Broking, document-store and form-library integration
W3Version pinning, edition matching and clause comparison
W4Consequence annotation, evaluation suite and document-cohort slices
W5Live renewals checked twice, and the two difference lists compared
W6Production validation, sign-off record and Agent Care handover
- Reading the band
- Week 5 runs the agent beside your own checkers on live renewals, not a retrospective sample. Each bar covers its own weeks only.
- At the end of W6
- Brokers are working from difference reports, and every check carries the name of the person who signed it and the wording they were shown.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Insurance AI agent
Check the wording before you tell a client nothing changed.
Show us an expiring policy, the renewal beside it and the check your people do today. We'll run one comparison end to end, show you every difference and every false one it raised, and write down what the report may never say.
- Typical delivery
- 4–6 weeks
- Deployment
- Your broking or policy stack
- Coverage advice
- A qualified broker or underwriter
- After launch
- Agent Care available
Nestack Agents · Renewal comparison & policy checking
AGT-INS-11 · Agent Care available after launch