Insurance AI agent · Commission reconciliation
Commission-Reconciliation AI Agent (Carrier Statements)
Match every carrier commission statement line to what the contract says was due, name the shortfalls, overpayments and missing statements, and package the difference — a person approves every adjustment.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Adjustments
- Approver signs
- After launch
- Agent Care
What this agent does
Finds the difference, moves no money
In
- 01
Take the statement as the carrier sent it — portal export, file, spreadsheet or download — and log what arrived.
- 02
Read the expected side from policy admin: policy, term, premium, and the producer of record on that date.
Reason
- 03
Match each paid line to a policy and a term, through rewrites, mid-term changes, reinstatements and cancellations.
- 04
Recompute what was owed from the commission schedule and the split agreement in force on that effective date.
- 05
Separate a shortfall from a chargeback, a timing difference and a period that never arrived at all.
Decide
- 06
A clean match inside the agreed tolerance closes, logged with the schedule version it cleared against.
- 07
Every difference, unmatched line, missing statement and licensing flag goes to a named approver.
Out
- 08
Package the difference per policy and term: paid, expected, and the clause the expectation rests on.
- 09
Retain the statement as received, the schedule and split version used, and who approved each adjustment.
Product statement
The agent matches, identifies and packages. Every payment, adjustment, write-off and dispute is a named approver's decision, and nothing is released to a producer whose licence and appointment for that business cannot be cleared. A clawback is a decision about someone's earnings — a person makes it, against the agreement it rests on, and it is not applied twice. This page covers commission owed on business already placed; it does not set rates, negotiate schedules or decide what a producer should be paid.
Example workflow
One statement, end to end
- 1Statement receivedA carrier statement as it arrived — portal export, file, spreadsheet or download — with the period it covers
- 2Expected side assembledPolicies, terms and premium from policy admin, with the schedule and split agreement in force on each effective date
- 3Lines matchedEach paid line to a policy and a term, through rewrites, mid-term changes, reinstatements and cancellations
- 4Differences classifiedShortfall, overpayment, chargeback, timing difference or a period that never came — each named and quantified
No human action required
Stages 1 to 4 run without a person in the loop — the statement is parsed, the expected side is assembled and every difference is classified before anyone opens the file. Nothing has been paid, adjusted or written off.
Closes with the versions it cleared.
Goes to a named approver, unresolved.
Commission or finance approver
Takes the pack with paid, expected and the clause behind the difference. Payments, adjustments, write-offs and disputes are theirs alone.
Approve · Adjust · Dispute
Approved — handed back
- 6Adjustment recordedOnly what a named approver signed — the ledger entry, the carrier dispute or the producer adjustment
- 7Outcome evaluatedMatch rate, discrepancy recall, false differences, licensing coverage and tolerance write-offs, by carrier and format
Approver changes
Every difference an approver reads differently is counted, in both directions.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Paying, adjusting or clawing back a producer's commission.
- Writing off a difference outside the signed tolerance.
- Releasing commission the licensing check cannot clear.
- Applying a chargeback or reversing a paid entry.
Automation boundary
Agent acts unaided
- ✓Parse the statement in whatever format the carrier sends.
- ✓Recompute what was owed from the schedule and split in force.
- ✓Classify each difference and name the clause behind it.
- ✓Assemble the pack and route it to a named approver.
Write actions run only inside the approval boundaries agreed during implementation. The agent may open a case and stage an entry. It may not move money, post an adjustment or release a commission, on any tier or in any configuration, and writes nothing off outside the signed tolerance.
- Opening or settling a dispute with a carrier.
- Changing a split, a schedule or a tolerance.
- Posting a commission adjustment to the ledger.
- Touching money held in a fiduciary or client account.
Example output
One statement line, annotated
Everything the agent proposes is attached to the statement line and the agreement it was recomputed from.
Reconciliation output · single statement line
Illustrative example- Statement
- Monthly, direct bill
- Policy
- Rewritten mid-term
- Producer
- Two-way split
- Difference
- Paid short, one term
- Confidence
- 92%
- Adjustment
- None — approver's
As received
The statement in the carrier's own format and period, and the policy, term and producer as policy admin holds them.
Checks cleared
- Matched to the rewritten term
- Split agreement in force
- Licence and appointment clear
Why it is short
The rewrite moved the term; the carrier paid on the schedule that applied before it.
What the score decides
Whether the match may close unattended — never whether the carrier owes it.
Action
ApproveAdjustDisputeValue
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 carrier statement
From portals, feeds and files
Recompute from the agreement in force
The commission schedule, the split and the override basis that applied on that effective date — not the ones filed today — read at reconciliation time and stamped with their version.
Close the lines that already agree
A line that matches its policy, its term and its schedule closes without anyone opening the statement, and carries the versions it cleared against.
Send the difference up with its clause
A shortfall, a double chargeback or a period that never came reaches a named approver with paid, expected and the agreement behind the expectation — never as a number on its own.
Build an evidence trail
Retain the statement as received, the schedule and split version used, the difference found, the approver's decision and what the carrier said back — on both paths.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Policy admin & agency management
- Guidewire
- Duck Creek
- Sapiens
- Applied Epic
- AMS360
Carrier statements & feeds
- Carrier portals
- statement files
- Commission download
- ACORD AL3 and XML
Ledger, payments & banking
- General ledger
- ERP
- payment rails
- Trust accounts
- fiduciary balances
Agent
Commission reconciliation
- Matches statements
- Classifies differences
- Routes for approval
Licensing & agreements
- NIPR and state records
- appointments
- Commission schedules
- split agreements
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 adjustment
Each control wraps the one inside it. A match clears every layer before it closes, and nothing that moves money clears them at all.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Decide whether the match may stand
- L3
- Keeps money from an uncleared producer
- L4
- Stops the quiet write-off
- L5 – L6
- Give the decision a name, and keep it
How Nestack evaluates it
Evaluate the match — and the difference nobody chased.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the pack an approver opens
- E1Match evaluationDid each paid line reach the right policy, term and producer?
- E2Recomputation evaluationWas the schedule, split and override basis the one in force that day?
- E3Discrepancy recallDid it find what a manual reconciliation of the same period found?
- E4Licensing coverageWas every payee on a split cleared before anything released?
- E5Slice evaluationHow does performance change by carrier, statement format and line?
- E6Business outcomeWhat was written off under tolerance, and what a carrier accepted.
Floor — the money that settles, and who it reached
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 →
01 · Statement intake
1 mode
- CR-01
Rows dropped in a new format
The export changed shape; the totals still balanced.
Stage reads
The statement as the carrier sent it, unchanged
02 · Matching
2 modes
- CR-02
Wrong term after a rewrite
Matched to a term the policy no longer has.
- CR-03
Chargeback matched twice
One cancellation clawed back on two statements.
Stage matches
Each paid line to a policy, term and producer
03 · Recomputation
2 modes
- CR-04
Split from a stale agreement
The producer agreement changed; the old split ran.
- CR-05
Override on the wrong basis
Bonus tier run on written, not earned premium.
Stage recomputes
What the schedule and split in force owed
04 · Difference
1 mode
- CR-06
Closed under tolerance
A real shortfall small enough that nobody read it.
Stage packages
The difference an approver opens, and its clause
05 · Release
1 mode
- CR-07
Cleared on a lapsed appointment
Licence checked at match, not again at release.
Stage tracks
What was approved, written off, disputed, by whom
- Sev-1 · a producer or a carrier is paid wrong
- Sev-2 · a real difference never surfaces
- Sev-3 · the pack is wrong, an approver catches it
Affected slices
A tolerance decides who absorbs the difference
A tolerance set as a flat amount rather than a proportion writes off hardest against the smallest producers, the newest agencies and anything billed in another currency. Nobody chose that; a threshold did. Nestack reports performance by slice, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Carriers that changed format | 5.9% | 3.6× | Review | |
| Cross-border and multi-currency | 4.1% | 2.5× | Review | |
| Split and sub-broker business | 3.1% | 1.9× | Watch | |
| Single-producer direct bill | 1.2% | 0.7× | Normal |
Bar: missed-difference lift vs. single-producer direct-bill baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
Carriers reject differences; the match is what failed
When a carrier sends one back, the fault is usually the term it was matched to or the schedule version it was recomputed from — not the amount.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Match rate, recall or write-off volume moves on one carrier.
- 02
Diagnose
Traced to the format, the match, the version pinned or the tolerance.
- 03
Improve
The parser, the matching rule or the tolerance is changed, re-signed by finance and version-linked.
- 04
Verify
Re-run against held-out statements, including every rejected difference.
- 05
Learn
The rejected difference becomes a regression case with its clause attached.
Learn → Detect
The return edge. Every cycle re-reads the tolerance itself — a threshold that quietly widened, or a write-off nobody put their name to, goes to finance before anything else ships.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, statement and agreement sourcing, matching and recomputation, evaluation, ledger integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Workflow discovery and automation-boundary definition.Week 1
- 02Carrier and statement-format inventory.Week 1, Week 2
- 03Policy-admin and agency-management APIs.Week 1, Week 2
- 04Statement parsing and change detection.Week 2
- 05Schedule, split and override modelling, pinned by date.Week 2, Week 3
- 06Policy, term and producer matching rules.Week 3
- 07Chargeback, reversal and timing logic.Week 3, Week 4
- 08Licensing and appointment validation.Week 3, Week 4
- 09Tolerance policy, register and approval routing.Week 4
- 10Match-rate and discrepancy-recall testing.Week 4, Week 5
- 11Ledger, payment and dispute-pack integration.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 carrier, one format | ProductionLive reconciliation | AdvancedMulti-carrier / multi-entity |
|---|---|---|---|
| Introduced at Pilot | |||
| Statement parsing and normalisation | ✓ | ✓ | ✓ |
| Policy, term and producer matching | ✓ | ✓ | ✓ |
| Difference classification and pack | ✓ | ✓ | ✓ |
| Draft mode — nothing is adjusted | ✓ | ✓ | ✓ |
| Baseline evaluation | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Schedule and split modelling by date | — | ✓ | ✓ |
| Licensing and appointment validation | — | ✓ | ✓ |
| Tolerance register and approval routing | — | ✓ | ✓ |
| Ledger and payment integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Overrides, contingents and bonus tiers | — | — | ✓ |
| Multi-currency and enterprise controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on the carriers and statement formats in scope, the policy-admin and ledger systems involved, schedule and split complexity, statement volume, the tolerance and approval controls 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.
You bring
We build with it
01The carriers, statements and periods in scope
→Workflow discovery and automation-boundary definition
Week 102Real statements in every format they arrive in
→Statement parsing and format-change detection
Week 203Access to policy admin or the agency system
→Policy, term and producer matching rules
Week 204Commission schedules, split and override agreements
→Schedule, split and override modelling, pinned by date
Week 305Your licensing and appointment source of record
→Licence and appointment validation before any release
Week 306A period you reconciled by hand, differences and all
→Evaluation suite, discrepancy recall and regression cases
Week 407The named approver, and the tolerance they will sign
→Tolerance register, approval routing 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 paid or adjusted before week 6; reconciliation runs on a period you have already closed.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Carriers, formats and who may approve an adjustment
W2Statement parsing, policy-admin access and the expected side
W3Schedules, splits, chargebacks and licensing validation
W4Evaluation suite, recall testing and the tolerance policy
W5Ledger and payment integration, in draft — nothing posts
W6First approved adjustments inside the signed tolerance, then handover
- Reading the band
- Bars span only the weeks their work is named in. Draft running in week 5 is real — the agent proposes adjustments and posts none.
- At the end of W6
- Adjustments have been posted only where a named approver signed them, and every release cleared the licensing check.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Insurance AI agent
Reconcile the statement before you argue about the money.
Show us a month of carrier statements in the formats they actually arrive in, the schedules and splits behind them, and who may approve an adjustment. We'll reconcile that month beside your own numbers and show you every line where the two of us disagree.
- Typical delivery
- 4–6 weeks
- Deployment
- Your agency and finance stack
- Adjustments
- Approved one by one
- After launch
- Agent Care available
Nestack Agents · Commission reconciliation & disputes
AGT-INS-12 · Agent Care available after launch
