Suggest the next procedural step on an open claim — the document to request, the deadline approaching — never a value, an offer, a reserve or a denial, which stay the adjuster's.
Take the claim as it stands now — file notes, documents received, tasks open, diary dates and the last activity.
02
Read the handling standard, the policy conditions and the statutory clocks that govern that claim in that state.
Reason
03
Compare the file with the procedure — what a reasonable investigation of this loss still needs, and what is done.
04
Name the gap: the document not requested, the statement not taken, the check not run, the specialist not involved.
05
Work out what falls due and when, from the clocks that started at notice and the diary the adjuster already keeps.
Decide
06
Rank the open steps by what is closest to a deadline and by what the file is missing, and say why for each one.
07
Say when a reserve review or a specialist referral is due, and never what the figure or the outcome should be.
Out
08
Put the steps on the adjuster's desktop, each one to accept, dismiss or ignore, with the file line behind it.
09
Retain the file state read, the suggestion made, what the adjuster did with it and any reason they gave.
→Product statement
The agent suggests procedure. Coverage, quantum, the reserve and anything adverse to the claimant are the adjuster's, and a suggestion may be dismissed without a reason and without consequence.
Example workflow
One open claim, end to end
AgentHuman
1Claim activityAn open claim that has moved — a document arrives, a task closes, a diary date falls due or a note is added
2File read as it standsNotes, documents, tasks, payments already made, diary entries and the handling standard for that loss and state
3Procedure comparedWhat a reasonable investigation of this loss still needs, set against what the file already holds
4Steps drafted and rankedEach step carries the gap it closes, the deadline behind it and the file line it was read from
No human action required
Stages 1 to 4 run without a person in the loop — the file read, the procedure comparison and the ranking are finished before the adjuster opens the claim. Nothing about the claim has changed.
5DecisionSplits on whether the step is procedural and the file is current
Procedural, and the file is current
Shown on the desktop as a suggestion.
Reads as a valuation, or the file is stale
Withheld, and the reason is logged.
Licensed adjuster
Accepts, dismisses or ignores each step. Coverage, quantum, the reserve and any decision adverse to the claimant remain theirs alone.
Accept · Dismiss · Ignore
Actioned — handed back▼
6Suggestions reach the desktopAccepted, dismissed or left alone — the claim itself moves only when the adjuster moves it
7Outcome evaluatedAcceptance and dismissal reasons, deadline adherence, missed-step recall and automation-bias indicators, by claim type and cohort
Dismissals
Every dismissed suggestion is counted; a reason is invited, never required.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required8 items
Suggesting a settlement value, a range or an offer.
Proposing a reserve figure, or a change to one.
Recommending a denial, a defence or a reservation of rights.
Deciding coverage, liability or quantum, in whole or part.
Automation boundaryAgent acts unaided
✓Read the open file and the handling standard that governs it.
✓Name the procedural step the file still needs, and what it closes.
✓Surface the statutory clock and the diary date that fall soonest.
✓Withhold anything that would read as a value, an offer or a denial.
Write actions run only inside the approval boundaries agreed during implementation. The agent suggests procedure; the adjuster decides the claim, and dismissing a suggestion is never incentivised or counted against them.
Saying anything to a claimant, insured or representative.
Closing a task, a diary date or a claim.
Calling a claim suspicious, or referring it to SIU.
Changing the suggestion set, its ranking or its thresholds.
Example output
One open claim, annotated
Every suggestion is attached to the file line and the handling standard it came from.
Guidance output · single open claimIllustrative example
Claim
File state
Clock
Top suggestion
Confidence
Value or reserve
Escape of water
Current, stamped
Statutory, open
Request the estimate
86%
None — never produced
As receivedThe file as it stood when it was read — notes, documents, open tasks and diary dates, with the time of the read.
Signals usedEstimate not on fileStatutory clock openLoss-type standard
Why this stepA reasonable investigation of this loss needs an estimate; the file holds none.
ActionAcceptDismissIgnore
What the score decidesHow high the step sits in the list. Dismissing it costs the adjuster nothing.
Value
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value landsValue 01 – 04
Every open claimFrom the claims desk and the diary
03Procedure & deadlines
Suggest against the handling standard, not the outcome
What a reasonable investigation of this loss type still needs, read against the file as it stands this morning — so a step is suggested because procedure calls for it, not because it would close the claim.
01Approved path
Bring the next step forward
The document not requested, the check not run and the specialist not yet involved are named while the claim is open, rather than found in a file review months later.
02Human review
Keep the number out of the guidance
Anything that would read as a value, an offer or a reserve figure is withheld before it reaches the desktop — guidance that shapes the settlement is a claims-practice problem long before it is a modelling one.
04Build an evidence trail
Retain the file state read, the step suggested, the standard behind it, what the adjuster did with it and any reason they gave — 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 adminGuidewire · Duck Creek · Sapiens Majesco · Claim, task and note APIs
Documents & evidenceEstimates and photos · repair records Medical records · portal and email intake
Diary, deadlines & desktopDiary and task queues · statutory clocks Adjuster desktop · work lists
Agent
Claims guidance & next-best-action
Reads the open file Suggests the next step Withholds any value
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the settlement
Each control wraps the one inside it. A suggestion clears every layer before it reaches the desktop, and everything about the claim itself sits outside all six.
L6 · Outermost — last line of defenceInward → L1 · closest to the model
L6Rollback / safe modeWithdraw guidance if acceptance, deadlines or slices degrade.Roll back
L5Claim-file recordSuggestion and outcome written where the file can be reconstructed.Record
L4Adjuster discretionA dismissal costs the adjuster nothing and is never incentivised.Protect
L3No-valuation ruleNo settlement value, offer, reserve figure or proxy for one.Withhold
L2Procedural scopeOnly steps the handling standard calls for, nothing about value.Scope
L1File-state freshnessBuilt on the file as it stands, or the suggestion is withheld.Refresh
Model coreSuggested step — the gap it closes, the standard behind it, the deadline it serves and a ranking score
L1 – L2Decide whether a step may be suggested
L3Keeps a value out of the guidance
L4Keeps the judgement with the adjuster
L5 – L6Keep the file whole and pull guidance back
How Nestack evaluates it
Evaluate what was suggested, what was ignored and what was missed.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the suggestion list an adjuster opens
Depth of coverage ▼
E1Final-output evaluationWas the suggested step the one the handling standard actually calls for?
E2Step-level evaluationWas it built on the file as it stood, or on a stale copy of it?
E3Tool evaluationDid it read the right claim, task, document and diary entry?
E4Adjuster responseWhat is accepted, what is dismissed, and for what stated reasons?
E5Slice evaluationHow does guidance quality change across claim types and claimant cohorts?
E6Business outcomeDeadline adherence, and the steps a reviewer later said were missed.
Floor — the step a file reviewer later said was missed
Failure modes
Where each failure originates in the agent
Seven failure modes plotted against the five stages of the agent lifecycle.
Agent lifecycleDirection of processing →
01 · File read1 mode
NB-01
Built on a stale file
The document arrived; the suggestion had not seen it.
Stage readsThe claim as it stands — notes, documents and tasks
02 · Procedure match1 mode
NB-02
Rare claim type, thin guidance
Where history is thin, the useful step is never offered.
Stage matchesThe handling standard for that loss type and state
03 · Drafting & ranking2 modes
NB-03
Suggestion carries a value
Procedural wording, but a number the adjuster anchors to.
NB-04
Deadline ranked below routine
A statutory clock sits under a document request.
Stage ranksWhich open step reaches the adjuster first, and why
04 · Desktop handover1 mode
NB-05
Guidance leans toward closure
Steps that close the claim outrank steps that complete it.
Stage suggestsThe suggestion shown, and what it may never say
05 · Drift / Version2 modes
NB-06
Suggestions followed unread
Accepted at a rate no judgement could account for.
NB-07
Cohort outcomes diverge
One claimant cohort is offered a shorter set of steps.
Stage tracksAcceptance, dismissal and suggestion mix over time
Sev-1 · guidance has shaped the settlementSev-2 · a required step is missed or lateSev-3 · the file moved, the suggestion did not
The gaps open widest where the agent has seen least
The missed steps concentrate in a few claim types, and a set that runs shorter for one group of claimants changes what those claims are investigated for and what they are paid. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregateIllustrative example
Slice
Failure rate
Lift
Lift vs. threshold
Status
Claim types with thin history
7.0%
3.7×
Review
Long-tail liability claims
5.5%
2.9×
Review
Reopened and supplemental claims
3.4%
1.8×
Watch
Repeat routine property claims
1.6%
0.8×
Normal
Bar: missed-step lift vs. routine property baseline · scale 0–4.0× · tick marks the 2.0× review threshold2 of 4 slices over threshold
Evidence-linked improvement
The suggestions nobody takes are the measurement
When adjusters dismiss the same step week after week, the guidance is wrong about the work rather than the adjusters being wrong about the claim.
Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn
01Detect
Acceptance, deadline adherence or missed-step recall moves in one slice.
02Diagnose
Traced to the file read, the standard matched, the ranking or the wording.
03Improve
The source, the ranking or the wording is changed, re-signed by the claims officer and version-linked.
04Verify
Re-run against held-out claims, including every step a reviewer said was missed.
05Learn
The missed step becomes a regression case with the file state that hid it.
Learn → DetectThe return edge. Every cycle also re-reads acceptance by adjuster and by claimant cohort — a set taken without question, or one quietly shorter for some claimants, is a finding.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, handling standards and file reads, gap detection and ranking, evaluation, desktop delivery, then production validation and handover.
WorkstreamWeek 1Week 2Week 3Week 4Week 5Week 6
01Workflow discovery and automation-boundary definition.
02Handling-standard and procedure sourcing by loss type.
03Claims-admin, diary and desktop APIs.
04File-state reads and freshness checks.
05Statutory-clock and diary-deadline mapping.
06Next-step drafting and gap detection.
07Ranking rules and deadline precedence.
08The no-valuation constraint and its test suite.
09Adjuster desktop, accept and dismiss capture.
10Acceptance and cohort-slice evaluation.
11Claim-file write-back and reconstruction trail.
12Deployment, documentation and Agent Care handover.
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallelFinal 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 queueProductionOne line, live on the desktopAdvancedMulti-line / multi-entity
Introduced at Pilot
Handling-standard procedure library✓✓✓
File-state reads and freshness checks✓✓✓
Next-step suggestions with the gap named✓✓✓
The no-valuation constraint and its tests✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Statutory clocks and diary deadlines—✓✓
Adjuster desktop, accept and dismiss capture—✓✓
Acceptance and automation-bias monitoring—✓✓
Cohort slicing and disparity reporting—✓✓
Introduced at Advanced
Reserve-review and referral timing prompts——✓
Multi-line and enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build priceConfirmed after discovery based on the lines and claim types in scope, the handling standards to be encoded, the claims, diary and desktop integrations, claim volume, the suggestion set agreed, and deployment requirements.
Separate from buildBuild 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 bringWe build with it
01The lines and claim types the guidance will cover→Workflow discovery and automation-boundary definitionWeek 1
02Your claim-handling standards and best-practice guides→Handling-standard and procedure sourcing by loss typeWeek 2
03Access to claims admin, the diary and the adjuster desktop→File-state reads, freshness checks and desktop deliveryWeek 2
04The statutory clocks and internal service dates you work to→Statutory-clock and diary-deadline mapping, and ranking rulesWeek 3
05Closed files where a reviewer said a step was missed→Evaluation suite, regression cases and failure-mode testingWeek 4
06Named adjusters, free to dismiss anything they disagree with→Adjuster desktop, accept and dismiss captureWeek 4
07Your written policy on who may value, reserve or deny→The no-valuation constraint, its tests and the claim-file trailWeeks 5–6
Nothing else is requiredDeployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Phases are drawn over the weeks they actually occupy. Nothing reaches an adjuster's desktop before week 6 — the suggestions go to a shadow queue first.
PhaseW1W2W3W4W5W6
DiscoveryW1
BuildW2 – W3
EvaluateW4 – W5
Pilot & LaunchW5 – W6
Week focusW1Lines, claim types and who may value, reserve or denyW2Handling standards and the procedures each loss type calls forW3File-state reads, gap detection and deadline mappingW4Ranking rules, evaluation suite, cohort slices and no-valuation testsW5Suggestions to a shadow queue, read by your claims leadsW6Live on the desktop, dismissals captured, then handover
Reading the bandBars span only the weeks their work is named in. The shadow queue in week 5 is real — no adjuster sees a suggestion.
At the end of W6Adjusters have worked live claims with suggestions on the desktop, and every value, reserve and denial in that period was theirs.
DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Insurance AI agent
Settle what the guidance may never say before it reaches a desktop.
Show us your claim-handling standards, how a diary date reaches an adjuster and who is allowed to value, reserve or deny. We'll run a week of your open claims through the suggestion set in shadow and show you every one we withheld for reading like a valuation.
