Main site

Travel & Hospitality AI agent · Expense

Corporate-Travel Policy & Expense AI Agent

Check each claim against your policy and the accountable-plan rules, surface duplicates, outliers and exceptions as signals for a named approver, and route what cannot be substantiated to payroll.

Typical delivery
4–6 weeks
Deployment
Your stack
Named approver
Signals only
After launch
Agent Care

What this agent does

A signal for an approver, not a verdict

In

  1. 01

    Ingesting the claim, the receipts, the trip record and the policy in force for that traveller.

  2. 02

    Normalising amounts, merchants, cost centres and trip legs, each carried with its own record.

Reason

  1. 03

    Testing each claim against business connection, substantiation and the return of excess.

  2. 04

    Reading a per diem as settling the amount only, and asking for time, place and business purpose.

  3. 05

    Watching the substantiation and return windows, and surfacing a claim before either closes.

Decide

  1. 06

    Raising duplicates, outliers and policy exceptions as signals for an approver to weigh.

  2. 07

    Routing what cannot be substantiated to the approver, and onward to payroll as a wage item.

Out

  1. 08

    Retaining the claim, the policy read against it, the substantiation, the edits and the approval.

  2. 09

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

Product statement

The agent checks and routes; a named approver decides, and the employer answers for how the arrangement itself is treated.

Example workflow

One claim, submission to approval

  1. 1Claim submittedExpense tool, card feed, receipt capture or a traveller's own submission
  2. 2Context assembledReceipts, the trip record, the cost centre and the policy in force for that traveller
  3. 3Claim reviewedPolicy read, substantiation, flags
  4. 4Controls appliedBusiness-connection checks, substantiation checks, window checks, flags and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is reimbursed at any of them — the agent is checking, and the approver's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold
High confidence

Goes to the named approver to release.

Low confidence

Adds a finance read first.

Approver releases

The claim is held with its policy read, its flagged lines and the confidence.

Approve · Query · Send to finance review

Approved — released for payment

  1. 6Expense systems updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedFlag outcomes, override rates, substantiation gaps and items reclassified after payment

Overrides

Every approver override is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Paying out a claim, or moving money against one.
    • Waiving substantiation for time, place or purpose.
    • Extending a substantiation or repayment deadline.
    • Calling a claim fraudulent, or naming a person.

  • Automation boundary

    Agent acts unaided

    • ✓Read each claim beside the policy in force for that traveller.
    • ✓Test business connection, substantiation and return of excess.
    • ✓Surface a duplicate or an outlier as a signal to weigh.
    • ✓Flag what an approver must decide, and hold the claim there.

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

    • Deciding whether a trip is safe for an employee.
    • Approving or refusing travel against a risk score.
    • Treating an unsubstantiated item as written off.
  • Changes to policy, thresholds or approval limits.

Example output

One claim, annotated

Everything the agent raises is attached to the claim and the policy it was read against.

Expense review output · single claim

Illustrative example
Claim
Two-night domestic trip
Flagged line
Meal claimed on a per diem, with no business purpose recorded against it
Amount claimed
$142.60
Source of record
Policy and trip record
Confidence
88%
Standing
Signal for the approver

As received

Taken from the claim, the receipts and the policy in force — nothing on this side is decided by the agent.

Records used

  • Policy in force
  • Trip and card record
  • Prior approver reads

Why this is flagged

A per diem settles the amount and nothing else — time, place and purpose are still missing.

What the score decides

Below the configured threshold the claim picks up a finance read before it reaches the approver.

Action

ApproveQuerySend to finance review

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 claim

From the claim and the policy

03Checking

Check against the policy

Read each claim against your written policy, the trip record and the accountable-plan requirements.

01Approved path

A flag is not a finding

Routine claims reach the approver already checked and already carrying what substantiates them.

02Human review

Send review to what money turns on

A duplicate, an outlier or an exception is a signal an approver weighs. The agent does not detect fraud and does not say that anyone committed any.

04

Build an evidence trail

The claim, the policy and substantiation behind it and the approver who released it stay on the report.

Integrations

Typical integrations

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

  • Expense platforms

    • SAP Concur
    • Expensify
    • Navan
    • Coupa
    • Brex

  • Card and payment feeds

    • Corporate card issuers
    • Bank and settlement feeds

  • Travel and booking data

    • Booking tools
    • agencies
    • Itinerary and trip records

Agent

Corporate travel policy & expense

  • Reads the claim
  • Checks the policy
  • Holds for approval

  • Finance and payroll

    • ERP
    • general ledger
    • Payroll 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 payment

The layers nest. What each one misses is named in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeFall back to policy checking alone when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, policy-rule and threshold configuration changes.Track
L4TraceabilityRecord the claim, the policy read, the flags, the edits and the approval time.Record
L3Approver releasesHold claims for the named approver; it governs release, not whether an approved claim is sound.Gate
L2Policy guardrailsTest claims against business connection, substantiation and the windows; a failure returns the claim.Restrict
L1Confidence thresholdsRoute low-confidence claims to a finance read before the approver sees them.Require review
Model coreClaim reviewed — the policy read, the substantiation, the flags and confidence
L1 – L2
Test whether a claim may stand
L3
Puts the release in an approver's hands
L4 – L5
Keep the claim and the substantiation behind it
L6
Reverts to policy checking when signals degrade

How Nestack evaluates it

Evaluate the review workflow — not only the decision on one claim.

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

Surface — the claim the approver reads

  1. E1Final-output evaluationDid the policy read match the policy actually in force for that traveller?
  2. E2Step-level evaluationDid the agent use the right trip record, cost centre and policy version?
  3. E3Tool evaluationDid it read and write the correct claim and the correct report?
  4. E4Confidence calibrationDo low-confidence claims actually attract more approver overrides?
  5. E5Slice evaluationHow does performance change across specific claim types?
  6. E6Business outcomeHow many claims were reclassified or corrected after payment?

Floor — the payroll position the employer carries

Failure modes

Where each failure originates in the agent

Seven failure modes, placed at the stage each one originates.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • CZ-03

      Wrong policy version read

      A claim is checked against a policy that no longer applies.

    Stage gathers

    Claims, receipts, trip records and the policy

  2. 02 · Reasoning

    2 modes

    • CZ-04

      Per diem read as complete

      The amount is taken to settle time, place and purpose too.

    • CZ-06

      Flag read as a finding

      A signal is written up as though it accused a person.

    Stage proposes

    Policy read, substantiation and confidence

  3. 03 · Tool / write

    2 modes

    • CZ-02

      Released without an approver

      A claim reaches payment with no person behind it.

    • CZ-05

      Substantiation window missed

      A claim passes the deadline before anyone is asked.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • CZ-01

      Risk note read as clearance

      A travel-risk line is taken as a decision that a trip is safe.

    Stage returns

    The claim the approver releases for payment

  5. 05 · Change / Version

    1 mode

    • CZ-07

      Silent policy regression

      A model or rule change loosens what the agent will pass.

    Stage tracks

    Model, prompt, policy rules and threshold config

  • Sev-1 · a claim released unapproved
  • Sev-2 · a wage event passes as expense
  • Sev-3 · policy data degrades, claim to review
See the complete Travel & Hospitality failure-mode catalogue

Affected slices

Overall quality can hide one bad cohort

Every number in this table is a ratio, and the denominator is what matters. Routine receipted claims sit at the bottom The cohorts that carry it are named, not averaged away..

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Per diem claims7.2%3.5×Review
Claims near a window edge4.5%2.2×Review
Flagged duplicates and outliers2.9%1.4×Watch
Routine receipted claims1.6%0.8×Normal

Bar: approver-override-rate lift vs. routine-receipted baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

A cycle ends in a standing test

A cycle is done when the miss has become a test the next release has to survive. That suite is what the next claim through review is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Approver-override rate rises in a claim slice.

  2. 02

    Diagnose

    The claims themselves come first — the receipts, the policy version read against them and the flags raised — and they are read until the cause narrows to one.

  3. 03

    Improve

    Version-stamp the change and attach the claims that exposed it.

  4. 04

    Verify

    The affected cases run again, and a fail stops the release.

  5. 05

    Learn

    It becomes a standing test, and the policy notes change with it.

  6. Learn → Detect

    The return edge. The next detection is measured against a longer suite.

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, review workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Expense and travel policy discovery and boundaries.Week 1
  2. 02Expense, card and trip source assessment.Week 1, Week 2
  3. 03Policy, substantiation and approval-limit rule mapping.Week 1, Week 2
  4. 04Claim ingestion and normalisation.Week 2
  5. 05Policy-check logic and record binding.Week 2, Week 3
  6. 06Confidence scoring and flag routing.Week 3
  7. 07Approver release workflow.Week 3, Week 4
  8. 08Expense and payroll integration.Week 5
  9. 09Substantiation and window cases.Week 4, Week 5
  10. 10Guardrails and approval controls.Week 4
  11. 11Claim-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 entity, one policyProductionProduction expense systemsAdvancedMultiple entities / policies
Introduced at Pilot
Checking against your policy✓✓✓
Approver release✓✓✓
Review-quality baseline✓✓✓
Introduced at Production
Reporting by cost centre—✓✓
Approval workflow in your systems—✓✓
Approved write-back—✓✓
Expense-system integration—✓✓
Introduced at Advanced
Multi-entity policy sets——✓
Multi-stage finance approvals——✓
High claim volume——✓
Multi-entity approval 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 written policy and approval limits

    Claim ingestion and policy mapping

    Week 1

  2. 02Claims you have already reviewed

    Review baseline, policy extraction and record binding

    Week 2

  3. 03Your substantiation rules and who approves

    Policy, substantiation and approval-limit rule mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Expense, card and trip-record assessment, then integration setup

    Week 2

  5. 05Claims you would not want reimbursed

    Substantiation cases and failure-mode testing

    Week 4

  6. 06What must reach an approver before money moves

    Confidence scoring, flag routing, guardrails and approval controls

    Week 3

  7. 07Named approvers to release claims

    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

Phases are drawn over the weeks they actually occupy, which is why week 5 doubles up rather than padding.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Expense workflow discovery, policy mapping and the automation boundary

W2Expense-system integration and the review baseline

W3Review workflow, confidence logic and approval controls

W4Substantiation cases, window guardrails and failure-mode testing

W5Payroll integration, pilot claims and targeted corrections

W6One reporting cycle reviewed under finance, then 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 last 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 · Travel & Hospitality AI agent

Build an expense agent that flags, and never accuses.

An approval outside the accountable-plan rules is not a policy exception; it is a payroll-tax event, and it attaches to everything paid under the arrangement rather than to the item that was waved through. So bring the three things that decide it — your written policy, the substantiation you actually require, and the named approver who releases money. This is a finance and tax control, not a spend-tracking tool.

Typical delivery
4–6 weeks
Deployment
Your expense stack
Named approver
Signals only
After launch
Agent Care available

Nestack Agents · Corporate travel and expense

AGT-TH-13 · Agent Care available after launch

Build an expense agent

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

About you
Your claims

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