Healthcare AI agent · Underpayment recovery
Underpayment-Recovery & Payer-Contract-Compliance AI Agent
Model what the contract said a paid claim should pay, read the remittance for what actually happened, explain each variance against the contract term behind it, and assemble the dispute a contract manager releases.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Manager release
- Post-payment
- After launch
- Agent Care
What this agent does
Models the variance, never files the dispute
In
- 01
A remittance posts against a claim already paid, from supported clearinghouse, payer portal or 835 sources.
- 02
The CARC and RARC codes are normalised, and each line carries forward the remittance it came from.
Reason
- 03
The contract and fee schedule are modelled for what the claim should have paid, carve-outs included.
- 04
The remittance is read for what happened — adjustments, take-backs and recoupments — against that rate.
- 05
Plan-funding status is established first: fully-insured, or self-funded and ERISA-governed.
Decide
- 06
A payer overpayment is flagged with the same discipline as an underpayment, and a silent denial too.
- 07
Every modelled variance, with the contract term behind it, routes to the named contract manager for release.
Out
- 08
The contract term, the remittance evidence, the manager's edits and the release stay on the claim.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent models the variance and assembles the dispute; the contract manager decides whether it is released, and provider standing and plan type are confirmed before anything moves.
Example workflow
One claim, remittance to release
- 1Remittance received835 remittance, clearinghouse feed, payer portal or lockbox scan
- 2Context assembledContract term, fee schedule, plan-funding status and remittance history, each with its source
- 3Variance modelledModelled rate, actual payment, contract term cited and confidence
- 4Controls appliedLookback-window checks, plan-type checks, standing checks and confidence threshold
No human action required
Stages 1 to 4 run unaided, and no dispute is assembled at any of them — the agent is modelling, and the manager's lane opens at the confidence gate.
Goes to the contract manager to release.
Adds a compliance read first.
Manager release
The claim is held with its modelled variance, its contract citation and the confidence.
Release · Edit · Send to compliance review
Released — dispute assembled
- 6Contract-manager system updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedVariance accuracy, manager edits, dispute outcomes and overpayment-reporting completeness
Manager edits
Every manager edit made at release is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Filing a dispute, appeal or demand with a payer.
- Adjusting, writing off or accepting a claim balance.
- Deciding that a payer's adjustment was improper.
- Filing the federal 60-day overpayment report or return.
Automation boundary
Agent acts unaided
- ✓Model the contract rate and read the remittance for what actually paid.
- ✓Carry every contract citation forward with the remittance line it explains.
- ✓Apply the state's configured lookback window and notice.
- ✓Flag a payer overpayment or a silent denial, and hold.
Any write happens inside the boundaries agreed at implementation, never ahead of the manager's release.
- Confirming provider standing where an anti-assignment clause is at issue.
- Applying one lookback window across states that set 180 days, 24 months and 365 days.
- Determining ERISA/self-funded status where the plan record is unclear.
- Changes to lookback rules, plan-type logic or release thresholds.
Example output
One claim, annotated
Everything the agent models is attached to the claim it was drawn from.
Variance output · single claim
Illustrative example- Claim
- Post-payment take-back, in-network
- Modelled variance
- Recoupment applied against the negotiated bundling rate — the contract's implant carve-out was not credited
- Payer & plan type
- Texas · state lookback rule
- Contract clause
- Signed contract on file
- Confidence
- 91%
- Manager status
- Plan confirmed fully-insured
As received
Taken from the claim record and the signed contract on file — nothing on this side is decided by the agent.
Evidence assembled
- Contract clause and rate
- 835 remittance codes
- State lookback rule
Why this reads as a variance
The remittance doesn't match the contract's carve-out term and a person still decides.
What the score decides
Below the configured threshold the claim picks up a compliance read before.
Action
ReleaseEditSend to compliance reviewValue
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 remittance posted
From the remittance feed
Model against the contract
Weigh the actual payment against the contract term, the fee schedule and the state's configured rules.
The duty runs both ways
A payer overpayment is surfaced with the same rigor as an underpayment — the model doesn't only look one direction.
Point the manager at what needs
Silent denials and low-confidence variances are marked, so the manager's read starts where the contract is ambiguous.
Build an evidence trail
The variance, the contract term it was read against and the manager who released it stay on the claim.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Contract & fee-schedule data
- Craneware
- Experian Health
- Contract repositories
- fee-schedule loads
Remittance & clearinghouse
- 835/837
- EDI feeds
- Waystar
- Availity
- Change Healthcare
Billing & PM systems
- Epic Resolute
- Cerner
- athenahealth
- other PM systems
Agent
Underpayment recovery & contract compliance
- Reads the remittance
- Models the variance
- Holds for release
Dispute & appeal tracking
- Contract-manager workflow tools
- Payer dispute portals
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 release
The layers wrap one another. What gets past them all is set out in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a variance may stand
- L3
- Puts the release in a manager's hands
- L4 – L5
- Keep the variance and the term behind it
- L6
- Holds the claim at flagged when signals degrade
How Nestack evaluates it
Evaluate the variance workflow — not only the assembled dispute.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the variance the manager sees
- E1Final-output evaluationDid every modelled variance match the contract term it was read against?
- E2Step-level evaluationDid the agent use the right contract, fee schedule and state lookback configuration?
- E3Tool evaluationDid it read the correct remittance, contract and claim record?
- E4Confidence calibrationDo low-confidence variances actually draw more manager edits?
- E5Slice evaluationHow does performance change across specific claim categories?
- E6Business outcomeHow many variances needed a manager edit or a correction after release?
Floor — the outcome the health system 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
- UR-03
Contract version behind the load
The fee-schedule load hasn't caught up with a signed contract amendment.
Stage gathers
Claim record, contract terms and remittance
02 · Reasoning
2 modes
- UR-04
Plan-type assumed, not confirmed
A self-funded ERISA claim is checked against a state prompt-pay statute that may not reach it.
- UR-06
Wrong lookback applied
A New York claim's 24-month window is checked against Texas's 180-day rule instead.
Stage proposes
The contract term, the variance and confidence
03 · Tool / write
2 modes
- UR-02
Overpayment finding dropped
The rules engine was tuned for underpayments only, and the 60-day federal clock runs unnoticed.
- UR-05
Standing not confirmed
A dispute drafts the provider as claimant before an anti-assignment clause is checked.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- UR-01
Silent denial read as routine
A take-back carries no CARC/RARC explanation and is filed as an ordinary contractual adjustment.
Stage returns
The dispute the manager releases
05 · Change / Version
1 mode
- UR-07
Silent lookback-rule drift
A state changes its recoupment-notice window and the per-state configuration isn't updated in time.
Stage tracks
Model, prompt, lookback and plan-type rules
- Sev-1 · released outside the boundary
- Sev-2 · wrong variance reaches the manager
- Sev-3 · remittance degrades, goes to review
Affected slices
One payer contract can hide inside a portfolio number
A portfolio recovery rate can look strong while a handful of claim categories absorb nearly all the rework. Nestack reports the variance-rework rate by claim category, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Post-payment take-backs on paid claims | 7.8% | 4.0× | Review | |
| Carve-outs, implants and stop-loss terms | 5.4% | 2.8× | Review | |
| Bundled and multi-procedure reductions | 3.5% | 1.8× | Watch | |
| Single-line fee-schedule claims | 1.9% | 0.7× | Normal |
Bar: variance-rework rate lift vs. the single-line fee-schedule baseline · scale 0–4.0× · tick at 2.0×
2 of 4 slices over threshold
Evidence-linked improvement
The loop does not close at recovery
A cycle shuts when the missed overpayment is a case the next release has to catch. That suite is what the next remittance read is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Variance-rework rate rises in a claim-category slice.
- 02
Diagnose
The take-back that arrived as an adjustment is traced to the remittance code, the contract term or the fee-schedule load behind it.
- 03
Improve
The change ships against a version, with the claims that exposed it attached.
- 04
Verify
A failing variance case holds the release until it clears.
- 05
Learn
The case joins the standing suite and the contract rules move with it.
Learn → Detect
The return edge. The next remittance read runs against a suite one case 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, variance workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Recovery-workflow discovery and boundary definition.Week 1
- 02Contract and remittance-source review.Week 1, Week 2
- 03Lookback and plan-type-config mapping and rule mapping.Week 1, Week 2
- 04Claim ingestion and normalisation.Week 2
- 05Variance modelling and contract binding.Week 2, Week 3
- 06Confidence scoring and flag routing.Week 3
- 07Manager release workflow.Week 3, Week 4
- 08Contract and billing-system integration.Week 5
- 09Variance and lookback cases.Week 4, Week 5
- 10Guardrails and release 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 payer, one contract | ProductionProduction billing-system integration | AdvancedMultiple payers / entities |
|---|---|---|---|
| Introduced at Pilot | |||
| Variance modelling and contract recommendations | ✓ | ✓ | ✓ |
| Manager release | ✓ | ✓ | ✓ |
| Variance-accuracy baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by payer | — | ✓ | ✓ |
| Release workflow in your systems | — | ✓ | ✓ |
| Approved dispute-packet write-back | — | ✓ | ✓ |
| Remittance integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-payer and multi-state rules | — | — | ✓ |
| Multi-stage manager approvals | — | — | ✓ |
| High claim volume | — | — | ✓ |
| Multi-contract recovery controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval 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 payer contracts, fee schedules and remittance access
→Contract ingestion and fee-schedule mapping
Week 102Representative claims across payer and category cohorts
→Variance-modelling baseline and remittance binding
Week 203Your state lookback rules and plan-type determinations
→Lookback-rule and plan-type-config mapping
Week 104Access to relevant APIs, feeds or exports
→Contract, remittance and billing-source assessment, then integration setup
Week 205Disputes you would not want filed
→Overpayment cases and the evaluation suite
Week 406What no variance may assume
→Confidence scoring, flag routing, guardrails and release controls
Week 307Named contract managers to release disputes
→Manager release 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
Each band sits on the weeks the work occupies, so week 5 runs evaluation and pilot together.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Recovery-workflow discovery, lookback mapping and the boundary
W2Contract and remittance integration, and the variance baseline
W3Variance modelling, confidence logic and release controls
W4Evaluation suite, guardrails and failure-mode testing
W5Billing-system integration, pilot claims and targeted corrections
W6One remittance cycle worked under the contract manager, 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 · Healthcare AI agent
Build an underpayment-recovery agent for claims that already paid — not pre-payment appeals, not aged AR.
Show us your contracts, your remittance feed and who releases a dispute. Next, we'll map the contract terms against your remittance data, set the state-by-state lookback rules, and name what only your contract manager releases.
- Typical delivery
- 4–6 weeks
- Deployment
- Your contracts & remittance
- Manager release
- Post-payment
- After launch
- Agent Care available
Nestack Agents · Underpayment recovery & contract compliance
AGT-HC-19 · Agent Care available after launch