Insurance AI agent · Group disability claims
Group-Benefits & Disability-Claims AI Agent (ERISA Procedure)
Take the enrolment and eligibility position, intake the disability claim, open the correct regulatory clock, assemble the medical and vocational evidence, and draft the determination and its notice for a qualified reviewer to decide.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Named reviewer
- Clock-first
- After launch
- Agent Care
What this agent does
Runs the clock and the file, never the decision
In
- 01
An employee enrols, and the agent records the eligibility position from the plan and the enrolment system.
- 02
A claim is filed, and the agent opens the clock from the date the plan received it, not the date it was assigned.
Reason
- 03
Medical and vocational evidence is assembled into the file, each item tied to its source.
- 04
The plan's own claims-and-appeals procedure is applied to the claim, not a generic timeline.
- 05
New evidence or a new rationale is surfaced to the claimant to answer before any final determination is drafted.
Decide
- 06
Missing evidence, an unaddressed treating-physician or SSA finding, and a clock at risk are flagged.
- 07
The drafted determination and notice are routed to the independent reviewer named for the claim.
Out
- 08
The claim, its clock, the evidence and the reviewer's decision stay on the file.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent drafts the determination and its notice; a qualified reviewer decides, and a licensed clinician decides anything turning on medical judgement.
Example workflow
One claim, intake to determination
- 1Claim receivedClaim form, plan documents, enrolment record or employer notice
- 2Clock openedFiling date, plan type, extension eligibility and the applicable claims procedure
- 3Evidence assembledMedical and vocational findings, treating-physician and SSA positions, and confidence
- 4Controls appliedClock-compliance checks, independence checks, disclosure-timing checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and no determination issues at any of them — the agent is assembling, and the reviewer's lane opens at the confidence gate.
Goes to the reviewer's queue as a completed file.
Routes for additional evidence before the file reaches the reviewer.
Independent review
The file is held with its evidence, its flagged gaps and the confidence.
Determine · Request more evidence · Escalate
Determined — released for notice
- 6Claims and plan systems updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedEvidence completeness, disclosure timing, clock compliance and post-determination appeals
Evidence returns
Every evidence return is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Denying, reducing or terminating a benefit.
- Deciding medical necessity or disability status.
- Setting an offset or the benefit amount.
- Closing a claim.
Automation boundary
Agent acts unaided
- ✓Intake the claim and open the clock the regulation requires.
- ✓Assemble the medical and vocational evidence, tied to its source.
- ✓Surface new evidence or rationale to the claimant before a final determination.
- ✓Flag what the file still needs, and hold the draft.
Any write happens inside the boundaries agreed at implementation, never ahead of the reviewer.
- Rescinding coverage outside the appeal procedure.
- Resolving a return-to-work or accommodation request.
- Marking a procedural miss de minimis on its own say-so.
- Changes to claims-procedure rules or state configuration.
Example output
One claim on the clock, annotated
Everything the agent assembles is attached to the source it came from.
Claim-file output · single claim
Illustrative example- Claim
- Own-occupation LTD claim
- Drafted finding
- States the evidence as insufficient, without addressing the treating physician's finding
- Clock status
- Day 22 of 45
- Evidence type
- Treating physician letter
- Confidence
- 87%
- Attribution
- Reviewer of record and licence ID
As received
Taken from the assembled evidence and the clock as filed — nothing on this side is decided by the agent.
Evidence used
- Treating physician note
- SSA determination
- Vocational assessment
Why this is flagged
The treating physician's finding is on the record and unaddressed.
What the score decides
Below the configured threshold before it reaches the approver.
Action
DetermineRequest more evidenceEscalateValue
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 claim
From the claim intake
Assemble against the clock
Draw on the assembled evidence and the plan's configured claims procedure.
The procedure is the substance
A clock that opens on the wrong date can send the claim to court before the plan ever gets an appeal.
Point the reviewer at what the file
Missing evidence, an unaddressed treating-physician finding and a clock nearing its deadline are flagged, so review starts where the deadline is closest.
Build an evidence trail
The claim, the clock it was filed against and the reviewer who decided stay on the file.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Claims & absence systems
- Claims/absence platforms
- Sedgwick
- Reed Group
- Matrix
Medical & vocational evidence
- IME and APS vendors
- ExamWorks
- MES
- IMX Medical
Plan & enrolment records
- Benefits admin
- payroll
- Workday
- ADP
- bswift
Agent
Group-benefits & disability claims
- Reads the file
- Assembles the evidence
- Holds for the reviewer
Notices & correspondence
- Print/mail
- e-delivery
- Document generation systems
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 the claim
The layers sit inside one another. What none of them catches is listed in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether the determination may proceed
- L3
- Puts the determination in the reviewer's hands
- L4 – L5
- Keep the claim and the clock behind it
- L6
- Narrows to intake and evidence assembly when signals degrade
How Nestack evaluates it
Evaluate the claims workflow — not only the final determination.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the notice the claimant sees
- E1Final-output evaluationDid the determination and notice match the evidence and the clock?
- E2Step-level evaluationDid the agent use the right procedure, evidence sources and state configuration?
- E3Tool evaluationDid it read and write the correct claim and the correct clock?
- E4Confidence calibrationDo low-confidence assemblies actually need more evidence?
- E5Slice evaluationHow does performance change across specific claim types?
- E6Business outcomeHow many claims needed more evidence or a later correction?
Floor — the outcome the plan answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- DB-03
Stale eligibility record
EOI or enrolment status is read from a record since superseded.
Stage gathers
Claim form, plan documents, enrolment record and history
02 · Reasoning
2 modes
- DB-04
Clock miscounted
The 45-day clock starts at assignment, not the date the plan received the claim.
- DB-06
Undisclosed new evidence
New evidence reaches the reviewer before the claimant sees it or can respond.
Stage proposes
Findings, clock status, flagged gaps and confidence
03 · Tool / write
2 modes
- DB-02
Rescission misrouted
A coverage rescission is processed as a correction, not the appeal procedure.
- DB-05
Clinical call automated
A medical-necessity point is treated as settled without a clinician's sign-off.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- DB-01
Treating-physician silence
The notice states a conclusion without addressing the treating physician's finding.
Stage returns
The notice the claimant receives and the plan
05 · Change / Version
1 mode
- DB-07
Silent procedure drift
A rule or model change widens what counts as an adequately disclosed rationale.
Stage tracks
Model, prompt, procedure rules and state configuration
- Sev-1 · determination outside boundary
- Sev-2 · unaddressed finding reaches notice
- Sev-3 · evidence degrades, routes to review
Affected slices
A compliant rate can hide one bad cohort
A blended turnaround rate across short-term and long-term claims can look compliant while a handful of claim cohorts absorb most of the procedural risk. Nestack reports the procedure-defect rate by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Claims where new evidence arrives late | 7.7% | 4.0× | Review | |
| Own-occupation to any-occupation changes | 5.5% | 2.9× | Review | |
| Claims with a contested eligibility date | 3.4% | 1.8× | Watch | |
| Short-term claims with clear evidence | 1.9% | 0.7× | Normal |
Bar: procedure-defect-rate lift vs. the short-term baseline · scale 0–4.0× · tick at 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
The cycle isn't done until the procedure holds
A cycle is done when the missed step has become a case the next release must pass. That suite is what the next determination made is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Procedure-defect rate rises in a claim slice.
- 02
Diagnose
The day the file sat while a form was ordered is traced back through the claim record until the cause narrows to one step.
- 03
Improve
Version-stamp the change and attach the claims that exposed it.
- 04
Verify
The affected cases run again, and a failure stops the release.
- 05
Learn
It becomes a standing test, and the procedure rules change with it.
Learn → Detect
The return edge. Detection next time runs against a suite one claim longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, claims workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Claims-procedure discovery and scope and boundary definition.Week 1
- 02Benefits and claims system assessment.Week 1, Week 2
- 03Procedure and state-configuration mapping and rule mapping.Week 1, Week 2
- 04Claim intake and clock normalisation.Week 2
- 05Evidence-assembly logic and source binding.Week 2, Week 3
- 06Confidence scoring and gap routing.Week 3
- 07Reviewer determination workflow.Week 3, Week 4
- 08Claims and benefits system integration.Week 5
- 09Deadline and notice cases.Week 4, Week 5
- 10Guardrails and review controls.Week 4
- 11Claim-trail instrumentation.Week 5, Week 6
- 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 tier | PilotOne plan, one line | ProductionProduction claims systems | AdvancedMultiple plans / states |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your procedure and rules | ✓ | ✓ | ✓ |
| Independent reviewer determination | ✓ | ✓ | ✓ |
| Procedure-compliance baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by benefit type | — | ✓ | ✓ |
| Reviewer workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Benefits-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-plan and multi-state rules | — | — | ✓ |
| Multi-stage reviewer approvals | — | — | ✓ |
| High claim volume | — | — | ✓ |
| Multi-plan disability controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, claim volume, review 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
01Your claim-file structure and plan documents
→Claim-file ingestion and clock mapping
Week 102Representative past claims, including denials
→Evidence-assembly baseline and clock logic
Week 203Your claims-procedure rules and state list
→Procedure and state-configuration mapping
Week 104Access to relevant APIs, feeds or exports
→Claims and benefits system assessment, then integration setup
Week 205Determinations you would not want issued
→Appeal cases and the evaluation suite
Week 406What no determination may skip
→Confidence scoring, gap routing, guardrails and review controls
Week 307Named reviewers to determine claims
→Reviewer determination workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The phases are drawn on the weeks they occupy, so week 5 genuinely doubles rather than padding the plan.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Claims-workflow discovery, procedure mapping and the automation boundary
W2Benefits-system integration and the assembly baseline
W3Assembly workflow, confidence logic and reviewer controls
W4Evaluation suite, deadline checks and failure-mode testing
W5Benefits-system integration, pilot claims and targeted corrections
W6One claim cycle worked under the disability lead, then Agent Care handover
- Reading the band
- A bar covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.
- At the end of W6
- The final checks clear on live claims and monitoring moves to Agent Care.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Insurance AI agent
Build a group-benefits and disability-claims agent around ERISA procedure, not a P&C timeline.
Show us your claims procedures and your plan documents. We'll map the intake-to-determination workflow and set the automation boundary; the clock, the evidence and the decision stay with your named reviewer.
- Typical delivery
- 4–6 weeks
- Deployment
- Your claims systems
- Independent review
- Clock-first
- After launch
- Agent Care available
Nestack Agents · Group-benefits & disability claims
AGT-INS-15 · Agent Care available after launch