Main site

Insurance AI agent · Claims settlement

Straight-Through Claims-Settlement AI Agent

Pay a narrow, pre-authorised band of low-value claims automatically once every coverage, quantum and screening check clears — the path only ever pays, and everything else goes to a licensed adjuster.

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

What this agent does

Settles the narrow band, refers everything else

In

  1. 01

    Take the claim as notified — policy, loss date, cover, claimant, payee, documents and the amount claimed.

  2. 02

    Read the policy as it stood on the loss date, including any endorsement, lapse or cancellation since.

Reason

  1. 03

    Test the claim against the eligibility band: line, claim type, cause of loss, value and evidence required.

  2. 04

    Compute the amount from the schedule, limits, sub-limits and excess in force — not from the sum claimed.

  3. 05

    Check duplicates, prior claims on the same loss, sanctions on the payee, and the bank details on file.

Decide

  1. 06

    Where every check clears and the amount sits under the ceiling, the claim pays without an adjuster.

  2. 07

    Anything short-paid, failing a check, or over the ceiling or the daily total goes to a licensed adjuster.

Out

  1. 08

    Pay the verified payee, apply the excess, and issue the settlement notice the policyholder receives.

  2. 09

    Retain the policy version, the checks that cleared, the figures used and the authority it paid under.

Product statement

The agent pays only inside a band, a per-claim ceiling and a daily aggregate limit a named claims officer signed. It never denies, never short-pays and never closes a claim against the claimant — those belong to a licensed adjuster, on every tier and in every configuration.

Example workflow

One claim, end to end

  1. 1Claim receivedA notified claim with the policy number, loss date, payee, documents and the amount claimed
  2. 2Policy read at the loss dateCover, endorsements, limits, sub-limits and excess as they stood, and whether the policy was in force
  3. 3Eligibility testedLine, claim type, cause of loss, value, evidence on file and the claim history behind it
  4. 4Quantum and screeningThe amount computed from the schedule and excess, then duplicate, fraud, sanctions and payee checks

No human action required

Stages 1 to 4 run without a person in the loop — the policy read, the eligibility test, the quantum and the screening all finish before the gate. Nothing has been paid.

5DecisionSplits on whether every check cleared and the amount sits under the ceiling
Every check cleared

Pays inside the ceiling and the daily total.

Any check failed, or over a limit

Goes to a licensed adjuster, unpaid.

Licensed adjuster

Takes the claim with the failed check named, the figures shown and the policy as read. Anything adverse to the claimant is theirs alone.

Pay · Amend · Investigate

Decided — handed back

  1. 6Payment released and notice sentOnly inside the per-claim ceiling, the daily total and the authority the claims officer signed
  2. 7Outcome evaluatedGate precision, overpayments, duplicates, payee failures, band drift and routing correctness, by claim type and channel

Adjuster changes

Every claim an adjuster pays differently from the computed figure is counted.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Declining a claim, in whole or in part.
    • Paying less than the claim asks for.
    • Applying an exclusion, condition or policy defence.
    • Settling a bodily-injury or liability claim.

  • Automation boundary

    Agent acts unaided

    • ✓Read the policy as it stood on the loss date.
    • ✓Test the claim against the band a named officer signed.
    • ✓Compute the amount from the schedule, limits and excess in force.
    • ✓Run duplicate, prior-claim, sanctions and payee checks before release.

    Write actions run only inside the approval boundaries agreed during implementation. Payment runs unattended only inside a band, a ceiling and a daily limit a named claims officer signed — never against the claimant. Recovery of a payment that should not have run is the carrier's, on that officer's authority, and counts against the band that allowed it.

    • Paying above the ceiling or the daily total.
    • Releasing payment after a bank-detail change.
    • Clearing a sanctions or fraud hit.
  • Changing the band, the ceiling or the limits.

Example output

One claim, annotated

Everything the agent computes is attached to the policy term and the check it came from.

Settlement output · single claim

Illustrative example
Claim type
Windscreen replacement
Policy at loss date
In force, unendorsed
Payee
Approved repairer
Settlement
$310.00
Confidence
96%
Deduction
Excess only, per schedule

As received

The claim as notified and the policy as it stood on the loss date, endorsements and lapse status included.

Checks cleared

  • Policy in force
  • No duplicate on file
  • Payee and bank verified

Why this figure

From the schedule, sub-limit and excess in force on the loss date, not the sum claimed.

What the score decides

Whether it may pay unattended, not what it pays. Below the threshold an adjuster takes it.

Action

PayAmendInvestigate

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 notified claim

From FNOL and the claims queue

03Policy & band

Apply the policy that was actually in force

Cover, endorsements, limits, sub-limits and excess as they stood on the loss date — read at settlement time and stamped with the version.

01Approved path

Let the small, verified claims clear

A claim inside the band, under the ceiling and clean on every check pays without an adjuster opening it. The band is narrow by design, and it is written down.

02Human review

Send anything adverse to a person

A failed check, a short payment, a policy defence or a value over the ceiling stops the path and reaches a licensed adjuster unpaid — and still inside the statutory clock, because referring a claim does not pause it.

04

Build an evidence trail

Retain the policy version read, the band and authority in force, the checks that cleared, the figures used and the adjuster's own decision — 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
    • Policy and claim APIs

  • Payments & finance

    • One Inc
    • VPay
    • ACH and card rails
    • Payment files
    • ledger reconciliation

  • Screening & verification

    • Sanctions and PEP lists
    • fraud scoring
    • Payee and bank-detail verification

Agent

Straight-through claims settlement

  • Tests eligibility
  • Computes quantum
  • Releases payment

  • Evidence & documents

    • FNOL attachments
    • photos and invoices
    • Repairer networks
    • claimant notices

  • 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 a payment

Each control wraps the one inside it. A claim clears every layer before money moves, and nothing adverse to the claimant clears them at all.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Kill switchA named claims officer stops the path in one action, at any hour.Stop
L5No-adverse ruleNo denial, no short payment, no policy defence — the path only pays.Withhold
L4Value limitsA per-claim ceiling, and a daily aggregate limit on the whole path.Cap
L3Payee screeningDuplicate, prior-claim, sanctions and bank-detail checks before release.Screen
L2In-force checkCover, endorsements and excess as they stood at the loss date.Verify
L1Eligibility gateOnly the claim types, causes and values inside the signed band.Gate
Model coreProposed settlement — the band it matched, the figures computed from the policy, and confidence
L1 – L2
Decide whether the claim may pay at all
L3
Decides whether this payee may be paid
L4
Caps one claim, and caps the day
L5 – L6
Keep it to paying, and stop it dead

How Nestack evaluates it

Evaluate what was paid — and what should never have paid.

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

Surface — the payment the claimant receives

  1. E1Final-output evaluationWas the amount right, and was the excess actually applied?
  2. E2Eligibility evaluationDid every claim that paid unattended belong inside the band?
  3. E3Tool evaluationDid it read the right policy version and pay the right payee?
  4. E4Screening evaluationWere duplicates, prior claims and sanctions caught before release?
  5. E5Slice evaluationHow does performance change across claim types, channels and cohorts?
  6. E6Business outcomeOverpayment and duplicate rate, and whether referred claims still paid on time.

Floor — the payments that should not have run

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 · Policy read

    2 modes

    • SC-01

      Paid on a policy not in force

      Cover had lapsed or been cancelled at the loss date.

    • SC-02

      Endorsement not read

      A mid-term change to cover or excess was missed.

    Stage reads

    Cover, endorsements, limits and excess at the loss date

  2. 02 · Eligibility

    1 mode

    • SC-03

      Stale data cleared the gate

      The band matched on figures already out of date.

    Stage tests

    Whether the claim belongs in the band that was signed

  3. 03 · Quantum & checks

    2 modes

    • SC-04

      Excess or sub-limit skipped

      Quantum taken from the wrong schedule or limit.

    • SC-05

      Duplicate of a settled claim

      The same loss already paid through another channel.

    Stage computes

    The amount, plus duplicate and sanctions checks

  4. 04 · Payment & notice

    1 mode

    • SC-06

      Wrong payee after a change

      Bank details changed shortly before release.

    Stage releases

    The money that leaves and the notice that follows it

  5. 05 · Band / Version

    1 mode

    • SC-07

      Band widens without notice

      A rule change admits a category nobody approved.

    Stage tracks

    Rule, threshold and model changes to the band itself

  • Sev-1 · money leaves and does not come back
  • Sev-2 · a real claim is paid the wrong amount
  • Sev-3 · paid twice, and recovery is possible
See the complete Insurance failure-mode catalogue

Affected slices

The band is only as narrow as its worst cohort

A band built from claim history, channel and geography decides who is paid in minutes and who waits for an adjuster. A few cohorts carry most of the wrong payments. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Policies endorsed mid-term6.4%3.2×Review
Payees with changed bank details4.4%2.2×Review
Claims just under the ceiling3.5%1.8×Watch
Repeat routine claims1.4%0.7×Normal

Bar: wrong-payment lift vs. repeat routine-claim baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Every wrong payment narrows the band

Money paid on a claim rarely comes back — absent fraud or an express recovery clause, a settlement stands. So each cycle changes what the band admits next.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Overpayment, duplicate or payee-failure rate moves in one slice.

  2. 02

    Diagnose

    Traced to the policy read, the gate, the quantum or a band change.

  3. 03

    Improve

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

  4. 04

    Verify

    Re-run against held-out claims, including every payment that had to be chased.

  5. 05

    Learn

    The wrong payment becomes a regression case and a gate the band must pass.

  6. Learn → Detect

    The return edge. Every cycle re-reads the band itself — a rule change that widened it without a signature is rolled back before the next claim pays.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, band and policy reads, quantum and screening, evaluation, payment integration, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and automation-boundary definition.Week 1
  2. 02Eligibility band, ceiling and daily-limit 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. 05Quantum from schedule, limits and excess.Week 2, Week 3
  6. 06Duplicate and prior-claim detection.Week 3
  7. 07Fraud and sanctions screening before release.Week 3, Week 4
  8. 08Payee and bank-detail verification.Week 3, Week 4
  9. 09Gate-precision and overpayment evaluation.Week 4, Week 5
  10. 10Payment integration, limits and the stop control.Week 5
  11. 11Settlement notices and finance reconciliation.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 claim type, shadow onlyProductionOne line, live paymentsAdvancedMulti-line / multi-entity
Introduced at Pilot
Eligibility band and gate✓✓✓
Policy-at-loss-date reads✓✓✓
Quantum from schedule and excess✓✓✓
Shadow mode — proposes, pays nothing✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Duplicate and prior-claim checks—✓✓
Sanctions and payee verification—✓✓
Live payment inside a signed limit—✓✓
Daily aggregate cap and stop control—✓✓
Introduced at Advanced
Multi-line and multi-entity bands——✓
Enterprise controls and audit packs——✓
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 payment systems involved, screening sources, claim volume, the authority and limits 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 and claim types you would pay without an adjuster

    Workflow discovery and automation-boundary definition

    Week 1

  2. 02A signed per-claim ceiling and daily aggregate limit

    Eligibility band, ceiling and daily-limit configuration

    Week 1

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

    Policy-at-loss-date reads and in-force checks

    Week 2

  4. 04Your schedules, sub-limits and excess rules

    Quantum computed from the limits actually in force

    Week 3

  5. 05Screening sources and your payee-verification rules

    Duplicate, prior-claim, sanctions and bank-detail checks

    Week 3

  6. 06Claims you paid wrongly, and the ones you had to chase

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  7. 07The named officer who holds the authority and the stop control

    Payment integration, limits and the stop control

    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 paid before week 6 — the path runs in shadow until the limits are proven on live claims.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Claim types, the band, the ceiling and who signs it

W2Policy-at-loss-date reads and endorsement handling

W3Quantum, duplicate, sanctions and payee verification

W4Evaluation suite, gate precision and failure-mode testing

W5Payment integration, limits and the stop control, in shadow

W6First live payments inside the signed limits, then handover

Reading the band
Bars span only the weeks their work is named in. Shadow running in week 5 is real — the path proposes and releases nothing.
At the end of W6
Live payments have run inside the signed ceiling and daily limit, and every adverse decision in that period was an adjuster's.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Insurance AI agent

Draw the band before you automate the payment.

Show us the claim types you would pay without an adjuster, the systems that hold the policy and the money, and who signs the authority. We'll draw the band, set the ceiling, the daily limit and the stop control, and show you everything that falls outside it.

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

Nestack Agents · Straight-through claims settlement

AGT-INS-08 · Agent Care available after launch

Build a straight-through settlement 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.