Main site

Accounting AI agent · AR & collections

Accounts-Receivable & Collections AI Agent (Cash Application & Dunning)

Match incoming remittances to open invoices, propose an explanation for every unapplied or short payment, run approved dunning contact, and route every dispute to the named credit officer who decides.

Typical delivery
4–6 weeks
Deployment
Your stack
Credit approval
Proposal-only
After launch
Agent Care

What this agent does

Matches the cash, not the credit decision

In

  1. 01

    Remittances arrive from bank feeds, lockbox files, payment portals and connected AR systems.

  2. 02

    Invoice references and payment identifiers are normalised before matching runs.

Reason

  1. 03

    Open invoices are matched against each remittance line by amount, date and reference.

  2. 04

    Configured cash-application rules and customer-specific matching logic are applied.

  3. 05

    Prior remittances and the customer's payment history inform an ambiguous match.

Decide

  1. 06

    Unapplied and short payments are flagged, each with a proposed explanation.

  2. 07

    Full-satisfaction language and disputed deductions are routed to the named credit officer, never auto-applied.

Out

  1. 08

    The remittance, the proposed match, the flags and every override are retained against the account.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

The agent proposes the match and the contact; the credit officer decides settlements, write-offs and credit limits, and the company remains the creditor of record.

Example workflow

One remittance, cash to contact

  1. 1Remittance receivedBank feed, lockbox file, payment portal or AR system API
  2. 2Context assembledOpen invoices, customer history, contact rules and prior remittances
  3. 3Match proposedProposed match, explanation, flags and confidence
  4. 4Controls appliedFull-satisfaction check, deduction check, per-state contact rules and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing posts or contacts anyone at any of them — the credit officer's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the credit officer to approve.

Low confidence

Adds a senior-credit read first.

Credit approval

The match is held with its evidence, its flags and the confidence.

Approve · Adjust · Send to credit manager

Approved — released to post

  1. 6Receivables system updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedMatch accuracy, override rate, dispute outcomes and days to resolve

Overrides

Every credit-officer override is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Applying or depositing a check marked paid in full.
    • Settling an account or agreeing a payoff.
    • Writing off a balance.
    • Releasing a lien, claim or guarantee.

  • Automation boundary

    Agent acts unaided

    • ✓Match incoming remittances to open invoices by amount and reference.
    • ✓Propose an explanation for every unapplied or short.
    • ✓Run approved dunning contact within configured.
    • ✓Route disputes and deductions to the named credit officer for the named owner.

    Any write happens inside the boundaries agreed at implementation, never ahead of approval.

    • Threatening legal or other collection action.
    • Deciding a credit limit, or the reasons behind it.
    • Reporting a balance to a credit bureau.
  • Ruling a deduction valid, invalid or fraudulent.

Example output

One remittance, annotated

Everything the agent proposes is attached to the account it came from.

Cash-application output · single remittance

Illustrative example
Customer
Regional distributor account
Remittance line
Wire remittance covering three invoices, one short
Amount
$18,240.00
Proposed match
Matched to INV-3381, 3382, 3384
Confidence
88%
Reason code
Pricing-discrepancy deduction, unconfirmed

As received

Taken from the remittance and the invoice ledger — nothing on this side is inferred.

Evidence used

  • Remittance advice text
  • Prior payment pattern
  • Open invoice balances

Why this match

The shortfall carries a deduction code the agent can't confirm.

What the score decides

Below the configured threshold the match routes to a senior-credit read before the credit.

Action

ApproveAdjustSend to credit manager

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 remittance

From the bank or lockbox feed

03Matching

Match against open invoices

Apply configured cash-application rules and the customer's payment history.

01Approved path

Chasing is not collecting

Routine remittances are matched and posted without a person touching them.

02Human review

Send review to the shortfalls

Short pays, deductions and full-satisfaction language are flagged, so the credit officer's read starts where risk concentrates.

04

Build an evidence trail

The balance, the evidence behind it and the credit officer who approved the step stay on the account.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • AR and invoicing systems

    • NetSuite
    • SAP
    • Oracle
    • Dynamics
    • Sage Intacct

  • Banking and remittance data

    • Bank feeds
    • lockbox files
    • Payment portals
    • EDI remittance

  • Customer records

    • Credit files
    • contact info
    • Contract and dunning history

Agent

Cash application & collections

  • Reads the ledger
  • Proposes the match
  • Holds for approval

  • Workflow

    • Email
    • ticketing
    • Approval 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 cash

Each control encloses the next. What the stack lets through is named in the map under it.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modePull the agent back to proposal-only when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, matching-rule and contact-policy changes.Track
L4TraceabilityRecord the remittance, the proposed match, the flags and every override.Record
L3Credit approvalHold matches for the named credit officer; it governs release, not whether the match is right.Gate
L2Policy guardrailsTest proposals against full-satisfaction, deduction and contact rules; a failure holds the item.Restrict
L1Confidence thresholdsRoute low-confidence matches to a senior-credit read before the credit officer sees them.Require review
Model coreMatch proposed — the match, explanation, flags and confidence
L1 – L2
Test whether a match may stand
L3
Puts the release in a credit officer's hands
L4 – L5
Keep the balance and the evidence behind it
L6
Drops the agent to statement-only contact when signals degrade

How Nestack evaluates it

Evaluate the cash-application workflow — not only the final match.

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

Surface — the match the customer sees

  1. E1Final-output evaluationDid the proposed match tie to the correct open invoices?
  2. E2Step-level evaluationDid the agent use the right ledger, contact rules and consent record?
  3. E3Tool evaluationDid it read and write the correct account and invoice?
  4. E4Confidence calibrationDo low-confidence matches actually attract more overrides?
  5. E5Slice evaluationHow does performance change across specific customer cohorts?
  6. E6Business outcomeHow many matches needed an override or a later correction?

Floor — the outcome the credit team answers for

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 · Retrieval

    1 mode

    • DU-03

      Superseded aging snapshot

      Invoice or customer history is read from a stale aging run.

    Stage gathers

    Remittance advice, invoice ledger and payer history

  2. 02 · Reasoning

    2 modes

    • DU-04

      Full-satisfaction blindness

      A short-pay check carrying full-satisfaction language is read as an ordinary partial.

    • DU-06

      Deduction mischaracterized

      A trade deduction is read as nonpayment and queued for dunning.

    Stage proposes

    The proposed match, explanation and confidence

  3. 03 · Tool / write

    2 modes

    • DU-02

      Low-confidence auto-apply

      Agent applies a match despite insufficient certainty.

    • DU-05

      Duplicate remittance posted

      Same remittance is applied to two invoices.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • DU-01

      Misapplied remittance

      Payment applied to the wrong open invoice.

    Stage returns

    The application the credit officer approves

  5. 05 · Change / Version

    1 mode

    • DU-07

      Silent contact-rule drift

      A model or rule change widens who the agent contacts, or how.

    Stage tracks

    Model, prompt, matching and contact rules

  • Sev-1 · a held action executes without
  • Sev-2 · a wrong match reaches the ledger
  • Sev-3 · source degrades, match routes to review
See the complete Accounting failure-mode catalogue

Affected slices

A healthy rate can hide one bad cohort

A healthy overall match rate can hide a handful of cohorts where unapplied cash concentrates. Nestack reports the unapplied-rate lift by slice, not only in total Nestack reports it by slice rather than in aggregate..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Short-paid with a deduction code6.7%3.5×Review
Remittance sent separately4.4%2.3×Review
Consolidated multi-invoice payments3.2%1.7×Watch
Single invoice, exact match1.9%0.8×Normal

Bar: unapplied-rate lift vs. the exact-match baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

The loop closes on a regression case

A cycle closes when the misapplied payment is a case the next release has to survive. That suite is what the next remittance posted is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Unapplied-rate rises in a remittance cohort.

  2. 02

    Diagnose

    The remittance that matched the wrong invoice is traced back through the rule and the record that produced it.

  3. 03

    Improve

    The fix ships against a version, with the accounts that exposed it attached.

  4. 04

    Verify

    Release is held until the affected cash-application cases pass again.

  5. 05

    Learn

    The case joins the permanent suite and the contact rules move with it.

  6. Learn → Detect

    The return edge. Detection next time 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, matching workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01AR workflow discovery and boundary setup and boundary definition.Week 1
  2. 02Bank, lockbox and AR-system assessment.Week 1, Week 2
  3. 03Contact-rule and state mini-FDCPA mapping and rule mapping.Week 1, Week 2
  4. 04Remittance ingestion and normalisation.Week 2
  5. 05Matching logic and explanation drafting.Week 2, Week 3
  6. 06Confidence scoring and dispute routing.Week 3
  7. 07Credit-officer approval workflow.Week 3, Week 4
  8. 08Receivables-system integration.Week 5
  9. 09Cash-application regression cases.Week 4, Week 5
  10. 10Guardrails and contact controls.Week 4
  11. 11Account-trail instrumentation.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 AR system, one entityProductionProduction receivables systemsAdvancedMultiple entities / systems
Introduced at Pilot
Matching and dunning recommendations✓✓✓
Credit approval✓✓✓
Match-accuracy baseline✓✓✓
Introduced at Production
Reporting by ageing band—✓✓
Approval workflow in your systems—✓✓
Approved cash posting—✓✓
Receivables-system integration—✓✓
Introduced at Advanced
Multi-state contact rules——✓
Multi-stage credit approvals——✓
High remittance volume——✓
Multi-entity receivables controls——✓
Build priceFrom $5,000From $8,000Custom 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.

  1. 01Your invoice and remittance data

    Remittance ingestion and matching setup

    Week 1

  2. 02Representative historical remittances

    Matching baseline, explanation drafting and confidence scoring

    Week 2

  3. 03Your approved contact rules and dunning cadence

    Contact-rule and automation-boundary mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Bank, lockbox and AR-system assessment, then integration setup

    Week 2

  5. 05Payments you would not want applied

    Short-pay cases and failure-mode testing

    Week 4

  6. 06What no contact may say on its own

    Confidence scoring, flag routing, guardrails and contact controls

    Week 3

  7. 07Named credit officers to review matches

    Credit-officer approval 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 covers the weeks the work really takes, which is why week 5 carries evaluation and pilot together.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1AR workflow discovery, contact-rule mapping and the boundary

W2Source integration and the matching baseline

W3Matching workflow, confidence logic and approval controls

W4Evaluation suite, guardrails and failure-mode testing

W5Bank integration, pilot remittances and targeted corrections

W6One ageing cycle run under the credit 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
Validation closes on live accounts, and Agent Care picks up monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Accounting AI agent

Build an AR and collections agent around your credit team's approval chain.

Show us your remittance sources, your contact rules and who signs off on exceptions. If short pays and deductions are still triaged by hand, we'll map the match logic, set the contact boundary and name who owns every dispute.

Typical delivery
4–6 weeks
Deployment
Your AR and banking stack
Credit approval
Proposal-only
After launch
Agent Care available

Nestack Agents · Cash application & collections

AGT-ACC-14 · Agent Care available after launch

Build an AR & collections agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your receivables

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.