Main site

Insurance AI agent · Claims triage

Claims-Triage & Adjudication-Support AI Agent

Score a new claim for severity and complexity, assemble the coverage position and the evidence behind it, and place it with an adjuster licensed and authorised for that claim — who decides.

Typical delivery
4–6 weeks
Deployment
Your stack
Adjudication
Adjuster only
After launch
Agent Care

What this agent does

Prepares and places the claim, decides nothing

In

  1. 01

    Take the claim as first notified — policy, loss date, cause, parties, reported injuries and whatever came with it.

  2. 02

    Read the policy as it stood on the loss date, endorsements included, and note what the notice left out.

Reason

  1. 03

    Score severity and complexity from the loss facts, the exposure and what the file does not yet hold.

  2. 04

    Assemble the coverage position — the grants, conditions, exclusions and endorsements that bear on this loss.

  3. 05

    Flag the injury, representation, subrogation and reserve indicators that change how a claim should be handled.

Decide

  1. 06

    Propose a segment and a track, and name the adjuster skill, licence and authority the claim needs.

  2. 07

    Send anything the score cannot separate, or the file cannot support, to a claims lead to place by hand.

Out

  1. 08

    Hand the adjuster the coverage position, the evidence cited, the open questions and the reserve indicators.

  2. 09

    Retain the policy version read, the signals scored, the segment, the desk it reached and every re-route.

Product statement

The agent scores, assembles and places. Coverage, quantum, the reserve and anything adverse to the claimant belong to a licensed adjuster with authority for that claim — on every tier and in every configuration.

Example workflow

One claim, end to end

  1. 1Claim notifiedA newly opened claim from FNOL, broker, portal or phone, carrying whatever the notice actually held
  2. 2Policy read at the loss dateCover, endorsements, limits, sub-limits and excess as they stood when the loss happened, and whether it was in force
  3. 3Severity and complexity scoredLoss facts, exposure, injury and representation indicators, and everything the file is still missing
  4. 4Coverage position assembledThe grants, conditions and exclusions that bear on this loss, each cited to the wording it came from

No human action required

Stages 1 to 4 run without a person in the loop — the policy read, the scoring and the coverage position are finished before an adjuster opens the file. Nothing has been decided.

5DecisionSplits on whether the score and the file support a track
Score and file agree

Goes to the matched adjuster, undecided.

Score unclear, or the notice too thin

Goes to a claims lead to place by hand.

Licensed adjuster or claims lead

Reads the claim itself, sets the track and the reserve, and decides coverage. Anything adverse to the claimant is theirs alone.

Accept · Re-route · Investigate

Placed — handed back

  1. 6Adjuster assigned, file handed overTo a skill, licence and authority level that matches the segment, inside the clock that started at notice
  2. 7Outcome evaluatedSegment agreement, severity calibration, coverage-issue recall, re-routes and time to the right adjuster, by channel and cohort

Re-routes

Every claim a person moves to another track or desk is counted.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Deciding coverage, in whole or in part.
    • Applying an exclusion, condition or policy defence.
    • Setting, raising or releasing a case reserve.
    • Denying, closing or withdrawing from a claim.

  • Automation boundary

    Agent acts unaided

    • ✓Read the policy as it stood on the loss date.
    • ✓Score severity and complexity, and show what the score rests on.
    • ✓Assemble the coverage position and cite each term to its wording.
    • ✓Name what the file is missing before an adjuster opens it.

    Write actions run only inside the approval boundaries agreed during implementation. The agent scores a claim and places it; a licensed adjuster with authority for that claim decides coverage, quantum and reserve.

    • Calling a claim suspicious, or referring it to SIU.
    • Placing a claim beyond an adjuster's licence or authority.
    • Telling a claimant anything about cover, liability or value.
  • Changing the segments, the score or the routing rules.

Example output

One new claim, annotated

Everything the agent scores is attached to the notice line or the policy term it came from.

Triage output · single claim

Illustrative example
Loss notified
Rear-end collision
Policy at loss date
In force, endorsed
Channel
Broker, partial
Segment
Injury track, complex
Confidence
88%
Coverage call
None — adjuster's

As received

The notice as it arrived and the policy as it stood on the loss date, endorsements and excess included.

Signals scored

  • Injury named at notice
  • Mid-term endorsement
  • Third party represented

Why this segment

Injury named at notice and a represented third party outrank the damage figure entirely.

What the score decides

Which adjuster opens it, not what they decide. The coverage issues reach them open.

Action

AcceptRe-routeInvestigate

Value

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 new claim

From FNOL and every channel

03Policy & segment

Score against the policy that was in force

Cover, endorsements, limits and excess as they stood on the loss date — read at triage time and stamped with the version, so the segment rests on the contract rather than on the notice alone.

01Approved path

Assemble the file before assignment

The policy read, the coverage position and the cited evidence are ready ahead of assignment, so the first hour of the claim is not spent finding out what is in it.

02Human review

Name the claim that cannot be placed

A claim the score cannot separate, or a notice too thin to place, reaches a claims lead instead of a track — and the clock keeps running, because triage does not pause the statutory timeframes.

04

Build an evidence trail

Retain the policy version read, the signals the score used, the segment proposed, the desk it reached and every re-route a person made — on both paths.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Claims & policy admin

    • Guidewire
    • Duck Creek
    • Sapiens
    • Majesco
    • Claim and policy APIs

  • FNOL & evidence capture

    • Phone and web FNOL
    • broker notice
    • Photos, estimates and reports

  • Adjuster workforce

    • Skill and licence registers
    • capacity
    • Authority levels
    • work queues

Agent

Claims triage & adjudication support

  • Scores severity
  • Assembles coverage
  • Places the claim

  • Reserving, litigation & SIU

    • Case reserving
    • litigation management
    • Referral queues
    • vendor panels

  • 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 model and an adjudication

Each control wraps the one inside it. A claim clears every layer before it is placed, and the adjudication itself sits outside all six.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeReturn triage to the claims desk if agreement or re-routes degrade.Roll back
L5Disparity monitoringSegment mix watched by channel, region and claimant cohort.Test
L4Licence and authorityA claim reaches only a desk licensed and authorised for it.Match
L3No-determination ruleNo coverage call, no reserve, no score that stands for one.Withhold
L2Evidence citationNo coverage issue is raised without the wording behind it.Cite
L1In-force policy readCover, endorsements and excess as they stood at the loss date.Verify
Model coreProposed triage — segment, severity and complexity, the coverage position and what the file is missing
L1 – L2
Decide whether the read may stand
L3
Keeps a coverage call out of the agent
L4
Puts the claim on a desk allowed to hold it
L5 – L6
Watch the mix and pull triage back

How Nestack evaluates it

Evaluate the placement — and what the adjuster did with it.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the segment and the file an adjuster opens

  1. E1Final-output evaluationDid the segment and severity match what an experienced adjuster set?
  2. E2Step-level evaluationWas the policy read the one in force on the loss date?
  3. E3Tool evaluationDid it read the right claim, policy and party, and write to the right file?
  4. E4Coverage-issue recallOf the issues an adjuster later raised, how many did it surface?
  5. E5Slice evaluationHow does the segment mix change across channels, regions and cohorts?
  6. E6Business outcomeRe-route rate after human review, and time to the right adjuster.

Floor — the claims that reached the wrong desk

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 →

  1. 01 · Notice intake

    1 mode

    • CL-01

      Thin notice reads as simple

      Complexity scored from a notice that is mostly blank.

    Stage gathers

    The claim as notified, and what the notice left out

  2. 02 · Policy read

    1 mode

    • CL-02

      Coverage issue not surfaced

      A mid-term endorsement changed cover and was missed.

    Stage reads

    Cover, endorsements and excess at the loss date

  3. 03 · Scoring

    2 modes

    • CL-03

      Severity under-scored

      A serious claim is placed on the fast track.

    • CL-04

      Litigation signal missed

      Injury or representation at notice does not move it.

    Stage scores

    Severity, complexity and the indicators behind them

  4. 04 · Routing & handover

    2 modes

    • CL-05

      Adjuster lacks the authority

      Placed with a desk unlicensed for that claim.

    • CL-06

      Reserve indicators not flagged

      The file gives no basis to set a case reserve.

    Stage places

    The desk the claim lands on and the file it carries

  5. 05 · Drift / Version

    1 mode

    • CL-07

      Triage drifts by channel

      One channel or region is quietly scored lower.

    Stage tracks

    Segment mix, rule changes and model versions over time

  • Sev-1 · the claim lands on the wrong desk
  • Sev-2 · the segment understates the claim
  • Sev-3 · the file is thin, review rebuilds it
See the complete Insurance failure-mode catalogue

Affected slices

The under-scoring clusters where the notice is thinnest

A score built on claim history, channel and geography decides who reaches an experienced adjuster within the hour and who waits a week. The mis-scoring sits in a few channels. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Sparse first-notice files6.1%3.4×Review
Phone-notified claims4.7%2.6×Review
Injury reported at notice3.2%1.8×Watch
Repeat routine motor claims1.4%0.8×Normal

Bar: mis-scoring lift vs. routine motor baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Adjusters correct the score by moving the claim

When an adjuster moves a claim to another track, the score was wrong that morning. That belongs in the evaluation, not a handling note nobody reads.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Segment agreement, coverage recall or re-route rate moves in one slice.

  2. 02

    Diagnose

    Traced to the notice, the policy read, the scoring or a routing rule.

  3. 03

    Improve

    The signal, threshold or routing rule is changed, re-signed by the claims officer and version-linked.

  4. 04

    Verify

    Re-run against held-out claims, including every one an adjuster re-routed.

  5. 05

    Learn

    The re-routed claim becomes a regression case with the missed signal marked.

  6. Learn → Detect

    The return edge. Every cycle also re-reads the segment mix by channel and cohort — a track that quietly stopped admitting one of them goes back to the claims officer who signed it.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, policy reads and notice extraction, scoring and coverage assembly, evaluation, adjuster matching, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Segment, track and severity-band sign-off.Week 1, Week 2
  3. 03Claims and policy-admin API assessment.Week 1, Week 2
  4. 04Policy-at-loss-date and endorsement reads.Week 2
  5. 05First-notice extraction and missing-data detection.Week 2, Week 3
  6. 06Severity and complexity scoring.Week 3
  7. 07Coverage-position assembly and citation.Week 3, Week 4
  8. 08Reserve, injury and litigation indicator flags.Week 4
  9. 09Adjuster skill, licence and authority matching.Week 4, Week 5
  10. 10Segment-agreement and disparity evaluation.Week 4, Week 5
  11. 11Claims-system handover and re-route capture.Week 5, Week 6
  12. 12Deployment, documentation 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 tierPilotOne line, shadow scoringProductionOne line, live routingAdvancedMulti-line / multi-entity
Introduced at Pilot
Severity and complexity scoring✓✓✓
Policy-at-loss-date reads✓✓✓
Coverage-position assembly✓✓✓
Shadow mode — scores, routes nothing✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Segment and track assignment—✓✓
Adjuster skill and licence matching—✓✓
Missing-evidence and reserve flags—✓✓
Disparity slicing by channel and cohort—✓✓
Introduced at Advanced
Litigation and referral-queue routing——✓
Multi-line and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on the lines and claim types in scope, the claims and workforce systems involved, notification channels, claim volume, the segments and authority rules agreed, 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.

  1. 01The lines, claim types and tracks you already run

    Workflow discovery and automation-boundary definition

    Week 1

  2. 02How you segment claims today, and who set the bands

    Segment, track and severity-band sign-off

    Week 1

  3. 03Access to claims and policy admin, endorsements included

    Policy-at-loss-date reads and coverage-position assembly

    Week 2

  4. 04Real first notices, including the thin ones

    First-notice extraction and missing-data detection

    Week 3

  5. 05Your adjuster skills, licences and authority levels

    Adjuster skill, licence and authority matching

    Week 4

  6. 06Claims an experienced adjuster re-segmented, and why

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  7. 07Your written policy on who may decide coverage

    Routing rules, the adjuster gate and the audit trail

    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. Nothing is routed live before week 6, and the scoring runs beside your existing triage until it is.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Lines, tracks, segments and who may decide coverage

W2Policy-at-loss-date reads and endorsement handling

W3First-notice extraction, scoring and coverage assembly

W4Adjuster matching, evaluation suite and disparity slices

W5Scoring beside your own triage, and agreement measured

W6First live placements, re-route capture, then handover

Reading the band
Bars span only the weeks their work is named in. The shadow run in week 5 is real — the agent scores and places nothing.
At the end of W6
Claims have been placed live from agent scores, and every coverage decision in that period was a licensed adjuster's.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Insurance AI agent

Agree the segments before you let a model assign them.

Show us how a claim reaches an adjuster today, the tracks you run and the licence and authority behind each desk. We'll score a week of your real notices beside your own triage and show you every claim where the two disagree.

Typical delivery
4–6 weeks
Deployment
Your claims and policy stack
Adjudication
A licensed adjuster, always
After launch
Agent Care available

Nestack Agents · Claims triage & adjudication support

AGT-INS-01 · Agent Care available after launch

Build a claims-triage agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your setup

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.