A small manufacturer reports water damage to its stockroom — seventy-four days after a pipe burst. In the meantime its policy has renewed with a higher deductible and a new water-damage endorsement. Which policy answers? The one in force today, which every screen in most claims systems shows, or the one in force on the day of the loss, which is the only one that matters?

Late notice is a coverage question, a prejudice question and sometimes a fraud question, and it is decided on the policy as it stood on the loss date. This post follows one late-reported claim through InsightCW: asking InsightUW for the policy at the loss date, the late-notice and fraud rules, the coverage agent's cautious position, and what the adjuster decides. The insured and amounts are illustrative.

Use case card
ActorsProperty adjuster (owner), coverage counsel if needed, claims manager, the insured and broker
TriggerA notice reported 74 days after the date of loss
LineCommercial property (the rules apply to every line)
Agents involvedIntake, coverage, fraud, correspondence, supervisor
Decisions and authorityThe coverage position on late notice (adjuster, with the manager on referral), whether the late report is a fraud indicator worth acting on
Systems touchedInsightUW (the policy at the loss date, not today's)
OutcomeThe right policy, the right deductible and endorsements, a coverage position that names the late notice and what it does and does not prejudice
What is measuredClaims handled on the wrong policy term (should be zero), days to coverage position on late notices, late-notice referrals by outcome

The wrong policy is the default

flowchart LR A["Late notice"] --> B["Policy screen shows
the current term"] B --> C["Today's deductible
and endorsements
applied"] C --> D["Coverage letter on
the wrong wording"] D --> E["Dispute; letter
withdrawn"] style C fill:#fdecec,stroke:#e57373 style E fill:#fff4e2,stroke:#f4a03a

Most claims systems copy a policy in at some point and show that copy. When a policy renews between the loss and the notice, the copy is the new term. Nobody chooses the wrong policy; the system chooses it for them.

flowchart LR A["Late notice"] --> B["UW: the policy
as of the loss date"] B --> C["Dated cache:
as_of = loss date,
UW's answer verbatim"] C --> D["Rules: late notice
60+ days; late report
indicator 30+ days"] D --> E["Coverage agent:
cautious position,
cited to the prior term"] E --> F{"Adjuster decides;
manager on referral"} style B fill:#efeafe,stroke:#7c6cf0 style F fill:#e8f8f4,stroke:#22c1a4,stroke-width:2px

How InsightCW asks UW

InsightCW does not own policies. When a notice names a policy number, it asks InsightUW one question: what did this policy say on this date? The call is service-to-service, read-only and signed. UW answers with the term, limits and aggregate, deductible or retention, coverage lines, the forms and endorsements on the quote schedule, and the trigger (occurrence, or claims-made with a retro date). CW keeps the answer verbatim with the loss date as its as_of, copies coverage onto the claim from it, and erodes the aggregate by what it has paid on the policy's other claims.

sequenceDiagram participant Ins as Insured participant CW as InsightCW participant UW as InsightUW participant Ag as Claims agents participant Adj as Adjuster participant Mgr as Claims manager Ins->>CW: Notice, loss 74 days ago CW->>UW: GET policy as_of the loss date (signed) UW-->>CW: The prior term: deductible, forms, endorsements CW->>CW: COV_LATE_NOTICE refers, FRD_LATE_REPORT indicator CW->>Ag: Autorun on conversion Ag-->>Adj: Coverage position (cautious), fraud weighing Adj->>Mgr: Referral: late notice Mgr->>CW: Decide position, ROR or confirm

Step by step

A late notice, from report to the coverage position Day numbers are from the notice. The loss was 74 days earlier. Amounts are illustrative.
1
Policy at the loss dateInsightUWDay 0

UW returns the prior term — the lower deductible and no water-damage endorsement. That is the policy the claim is handled on. If UW cannot be reached, CW uses whatever policy it already holds and says on the file that the answer did not come from UW.

2
Late-notice referralCOV_LATE_NOTICEDay 0

Late notice of 60 days or more refers the coverage position. The policy-in-force rule (COV_NOT_IN_FORCE) stays quiet: the loss date is inside the prior term.

3
Late-report indicatorFRD_LATE_REPORTDay 0

A notice 30 or more days after the loss raises a medium fraud indicator. It is an indicator, not an accusation; the score stays below the SIU threshold.

4
Coverage positioncoverage_agentDay 0

Water damage covered under the prior term's wording, with a reservation of rights on late notice and prejudice. It cites the prior term's forms, not the current ones. The position can only move towards caution, and a denial would need a form number on the policy at the loss date.

5
Fraud weighingfraud_agentDay 0

The late report's suspicious reading (the damage grew, or was not sudden) and its innocent one (the stockroom is rarely used; the owner was away). Recommendation: monitor, not refer.

6
Information requestintake_agentDay 1

Asks for when the leak was discovered, by whom, and photographs or invoices from the time — the facts that decide prejudice. Drafted from the template, held for the adjuster.

7
Position takenAdjuster and managerDay 6

With the discovery date established, the manager agrees the reservation of rights. The coverage clock is met and the decision record keeps the position, the prior-term form it cited and the referral it answered.

What the screens show

The Claim Workspace for a general liability claim in suit, on demo data
The Claim Workspace on a general liability slip-and-fall in suit (demo data): the hero with the stage strip and the next step (Confirm coverage), the agents’ narrative with Analyze and “What would they do?”, and the key facts.
What the agents found, a card per agent with model, confidence, findings and sources, on demo data
What the agents found, on a general liability claim (demo data). Each card shows the model that reasoned (Claude Haiku 5.5), its confidence, findings with a chip per source, open questions, tokens and cost, and “Run this agent again”. Every agent's card has the same shape.
The claims copilot answering a question about a claim, with numbered sources, on demo data
The claims copilot on a general liability claim (demo data) answering “Is the reserve adequate?”: where the reserve stands and why it looks short, each figure with a numbered source.

What the agents said, and what each answer rests on

AgentWhat it preparedWhat it citesWho decides
IntakeAsk for the discovery date and contemporaneous evidenceNotice date, loss date, documents on fileAdjuster sends
CoverageCovered under the prior term; reservation of rights on late noticeUW's answer as of the loss date; COV_LATE_NOTICEAdjuster; manager on referral
FraudMonitor, do not referFRD_LATE_REPORT, the score, the facts on fileAdjuster
CorrespondenceAcknowledgement due on the 24-hour clockThe acknowledgement clockAdjuster sends

Edge cases

What happensWhat InsightCW does
The loss date is outside every termCOV_NOT_IN_FORCE refers; the coverage agent cannot call it covered.
The loss date changes after investigationThe policy is asked for again as of the new date; every cached answer carries the date it was asked for.
Loss within 30 days of inceptionCOV_EARLY_LOSS refers, and the recent-policy fraud indicator rises.
A claims-made policyThe question becomes when the claim was first made and the retro date, as in the D&O post.

The decision record for this claim

SectionHitOutcome
CoverageWater damage covered under the prior term (policy at the loss date from UW)warn
CoverageCOV_LATE_NOTICE: late notice of 60 days or morerefer
FraudScore 15: FRD_LATE_REPORT (30 days or more after the loss)warn
ClocksAcknowledgement met; coverage decision clock runningwarn
AgentCoverage: reservation of rights on late notice and prejudice (source: agent)warn

Verdict refer. The record names the policy term the claim was handled on, which is the first thing anyone will ask if the position is disputed.

Ask the copilot

Ask the copilotLookups it callsWhat the answer covers
“Which policy term applies?”policy_and_coverageThe term in force on the loss date, its deductible and forms, and that the answer came from UW.
“How late was this reported?”claim_overview, decision_recordLoss and report dates, the days between, and the rules that fired.
“What do we still need to decide prejudice?”notes, documents, diary_and_clocksWhat is on file and what the information request asked for.

Tuning it for your book

  • The late-notice threshold. 60 days is the default for the coverage referral and 30 days for the fraud indicator. Some lines and states need different numbers.
  • The UW connection. Configure the signed lookup so CW always asks UW. A fallback policy is labelled as such; aim never to need it.

On the desk

You handle the claim on the policy that was in force, without looking for it. The late notice is a referral with the facts, not a surprise.

For claims IT

GET /uw_service/policies/{number}?as_of=YYYY-MM-DD, signed; the answer is cached verbatim in cw_policy with source_system and as_of.

For claims leadership

Claims handled on the wrong policy term should be zero. It is a measurable, auditable number.

Variations

A late notice on a claims-made policy

On claims-made wording, late notice is usually a reporting-provision question: was the claim reported within the policy period or the extended reporting period? The management-liability line facts and checks carry it, as in the D&O post.

A loss just after inception

COV_EARLY_LOSS refers a loss within 30 days of inception, and the recent-policy fraud indicator rises. Most such losses are honest; the referral exists so the inception date is checked against the loss date by a person.

No policy number on the notice

Intake cannot ask UW for a policy it cannot identify. The intake agent prepares an information request; the coverage agent says it has no policy to cite and prepares nothing.

Rolling it out

  1. Configure the UW lookup with its signed secret, and confirm it answers for a sample of policies that have renewed.
  2. Agree the thresholds for late notice (60 days) and the late-report indicator (30 days) with coverage counsel and SIU.
  3. Measure claims handled on the wrong term (zero) and days to coverage position on late notices.

Questions we get asked

What if UW is down when the notice arrives?

CW uses the policy it already holds — a seeded or manual policy — and says on the file that the answer did not come from UW. It asks again when it can.

Is a late report treated as fraud?

It raises a medium indicator, which on its own stays below the SIU threshold. The fraud agent weighs it and usually recommends monitoring.

Can an auditor replay what happened?

Yes. Every Analyze, stage action and payment hold writes a run to the decision record, with the hits in order, the rule and facts behind each, and the verdict. Every agent step records its model, tokens, cost and confidence, and a Lense trace when tracing is configured.

Which model do the agents use, and what if it is unavailable?

By default the agent steps use Claude Haiku 5.5 through the platform's model client (on Bedrock where a region is configured), with a daily budget per agent; over budget, the fast model answers. If no model is configured, or a call fails, or the answer does not parse, the agents' rule-based findings stand and the step says the model did not run. Nothing in the claim waits on a model.

What the person still decides

  • Whether the late notice prejudiced the carrier. A legal and factual judgement.
  • The coverage position and the letter.
  • Whether the late report means anything beyond lateness.

Keep reading