Accounting AI agent · Indirect tax

Sales-Tax & Compliance-Filing AI Agent

Determine taxability, prepare and reconcile indirect-tax returns and assemble the filing package — submission and payment stay with an authorised person, with evaluation, audit trails and guardrails built in.

Typical delivery
4–6 weeks
Deployment
Your stack
Human review
Before filing
After launch
Agent Care

What this agent does

Prepares the return, holds the submission

In

  1. 01

    Ingest sales and purchase transactions from supported ERP, billing, e-commerce and marketplace sources.

  2. 02

    Normalise each line into customer, ship-from and ship-to, product or service code and date.

Reason

  1. 03

    Determine taxability and rate for the jurisdiction a transaction sources to, on the date it sources.

  2. 04

    Apply the registration footprint, exemption certificates and the client's agreed taxability positions.

  3. 05

    Prepare each return and reconcile it to the transaction detail and the general ledger.

Decide

  1. 06

    Detect tie-out variances, certificate gaps, nexus thresholds in reach and calendar changes.

  2. 07

    Route variances, unregistered jurisdictions and low-confidence determinations to human review.

Out

  1. 08

    Retain the workpaper trail — source detail, rate applied, tie-out and reviewer decision.

  2. 09

    Assemble the filing package and hold it; submission and remittance need explicit authorisation.

Product statement

The agent prepares and reconciles inside the approval boundaries agreed during implementation; an authorised person submits the return and releases the money.

Example workflow

One filing period, end to end

  1. 1Period drawnERP, billing, e-commerce and marketplace transactions
  2. 2Positions resolvedRegistration footprint, sourcing, taxability, certificates and the rates in force on each date
  3. 3Return preparedLiability by jurisdiction, schedules and the ledger tie-out
  4. 4Controls appliedTie-out variance, certificate coverage, nexus and calendar checks, and the confidence threshold

No human action required

Stages 1 to 4 run without a person in the loop — nothing has been filed or paid yet, so the lane stays empty until the gate.

5DecisionSplits on the tie-out and confidence gate
Ties out

Goes forward for authorisation.

Variance or gap

Held for the indirect-tax reviewer.

Reviewer authorisation

The package is held with its workpapers, tie-out, certificate coverage and confidence.

Authorise · Adjust · Escalate

Authorised — handed back

  1. 6Filing package handed overAn authorised person submits and pays; the agent attaches the confirmation to the workpapers
  2. 7Outcome evaluatedTie-out variance, taxability accuracy, certificate coverage, on-time filing, notices and amendments

Adjustments

Adjustments and escalations made in review are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Submitting a return to any filing portal.
    • Releasing a remittance or approving a payment file.
    • Registering the business in a new jurisdiction.
    • Amended returns and refund claims.
  • Automation boundary

    Agent acts unaided

    • ✓Determine taxability, sourcing and rate for each transaction.
    • ✓Apply the registration footprint, certificates and agreed positions.
    • ✓Prepare returns and reconcile them to the general ledger.
    • ✓Route variances and coverage gaps to review with the workpapers.

    Returns are submitted and money released only inside the approval boundaries agreed during implementation.

    • Responses to notices and assessments.
    • New taxability positions for a product or service.
    • Accepting a certificate the agent cannot validate.
  • Changes to rate sources, calendars or thresholds.

Example output

One prepared return, annotated

Everything the agent proposes is attached to the period it was built from.

Prepared return · single period

Illustrative example
Jurisdictions
State, home-rule city
Taxable sales
$318,400.00
Period
Monthly
Prepared return
Ready to authorise
Confidence
91%
Ledger tie-out
Within tolerance

As received

Taken from the transaction detail for the closed period — nothing here is restated to make it balance.

Evidence used

  • Rate in force at the date
  • Certificate on file
  • Return-to-ledger tie-out

Why two schedules

The home-rule city taxes a category the state exempts, so each base is prepared apart.

What the score decides

Below the configured threshold the package is held for review instead of going forward for authorisation.

Action

AuthoriseAdjustEscalate

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

All taxable transactions

From billing and the ERP

03Determination

Apply your own footprint

Use the registration footprint, exemption certificates and the taxability positions already agreed.

01Approved path

Cut the preparation and tie-out work

Routine periods come back prepared, reconciled and evidenced instead of being rebuilt by hand each month.

02Human review

Put reviewers on the variances

Unreconciled periods, certificate gaps and new jurisdictions reach a reviewer instead of every return doing so.

04

Build an evidence trail

Retain the source detail, rate applied, tie-out, confidence, evaluator result and reviewer decision — on both paths.

Integrations

Typical integrations

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

  • Transaction sources

    • ERP
    • billing
    • e-commerce
    • POS
    • Marketplace feeds
  • Tax engines & rate data

    • Avalara
    • Vertex
    • Sovos
    • Rate and boundary files
  • Certificates & documents

    • Certificate systems
    • customer master
    • Document management
    • notice files

Agent

Sales-tax & compliance filing

  • Determines taxability
  • Prepares and ties out
  • Holds for authorisation
  • Filing & remittance

    • Filing portals
    • e-file channels
    • Banking for remittance
  • 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 a filed return

Each control wraps the one inside it. A return clears every layer before anyone is asked to sign it.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeRestrict automation if evaluations or production signals degrade.Roll back
L5Version monitoringTrack rate and rule updates, model, prompt and configuration changes.Track
L4TraceabilityRecord source detail, rate applied, tie-out, decision and adjustment.Record
L3Filing authorityDefine who may submit a return and release the remittance.Gate
L2Scope guardrailsRestrict returns to registered jurisdictions, periods and agreed positions.Restrict
L1Tie-out checksPrepared liability must reconcile to the detail and the ledger.Reconcile
Model coreReturn prepared — liability by jurisdiction, schedules, tie-out and confidence
L1 – L2
Decide whether the return may stand
L3
Decides who may submit and pay
L4 – L5
Keep the record and the rates current
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the full workflow — not only the filed number.

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

Surface — the return the reviewer signs

  1. E1Final-output evaluationWas the liability right for that jurisdiction and period?
  2. E2Determination evaluationWere taxability, sourcing and rate correct per transaction?
  3. E3Reconciliation evaluationDoes the return tie to the detail and to the ledger?
  4. E4Escalation calibrationAre variances and certificate gaps actually held back?
  5. E5Slice evaluationHow does performance change across specific filing cohorts?
  6. E6Business outcomeHow many returns were amended, late or drew a notice?

Floor — the outcome the client pays 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

    • ST-01

      Rate applied from the wrong date

      A rate change is read from the wrong quarter.

    Stage gathers

    Transactions, sourcing data, certificates and rates

  2. 02 · Determination

    2 modes

    • ST-02

      Sourced to the wrong jurisdiction

      Origin used where the state sources to delivery.

    • ST-03

      Exemption claimed without support

      A missing or expired certificate reads as valid.

    Stage proposes

    Taxability, rate, jurisdiction and confidence

  3. 03 · Tool / write

    2 modes

    • ST-04

      Filed or paid without authorisation

      Submitted or paid before anyone authorised it.

    • ST-05

      Filed to the wrong period

      The return lands in a period already closed.

    Stage writes

    Only where filing authority and approval policy allow

  4. 04 · Output

    1 mode

    • ST-06

      Return does not tie to the ledger

      Prepared liability and the ledger disagree.

    Stage returns

    The prepared return and its tie-out to the ledger

  5. 05 · Change / Version

    1 mode

    • ST-07

      Nexus or calendar drift

      A crossed threshold or a new frequency is missed.

    Stage tracks

    Rate updates, calendars, model and configuration

  • Sev-1 · a filing obligation is breached
  • Sev-2 · wrong liability reaches the return
  • Sev-3 · tie-out fails, package routes to review
See the complete Accounting failure-mode catalogue

Affected slices

One jurisdiction mix can carry the whole error rate

Single-state monthly filings behave predictably; home-rule cities, exempt sales and newly registered states are where determination and tie-out errors concentrate. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Home-rule jurisdictions5.2%3.3×Review
Exempt and resale sales3.7%2.4×Review
Newly registered states2.8%1.8×Watch
Single-state monthly filings1.0%0.6×Normal

Bar: lift vs. single-state baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

Rates and certificates go stale on a date

Each cycle ends with a rate source re-dated, a certificate chased or a tie-out rule tightened — and the next period is measured against that.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Tie-out variance or notices rise in a cohort.

  2. 02

    Diagnose

    Traced to a rate source, a sourcing rule, a certificate or the feed.

  3. 03

    Improve

    The fix is approved, version-linked and dated from the period it applies.

  4. 04

    Verify

    Affected periods are prepared and tied again.

  5. 05

    Learn

    That period becomes a standing case in the regression calendar.

  6. Learn → Detect

    The return edge. The next period starts with one more dated rule and one more re-tied case.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, data, determination logic, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and filing-authority definition.Week 1
  2. 02Registration footprint and filing calendars.Week 1, Week 2
  3. 03Source-system and tax-engine assessment.Week 1, Week 2
  4. 04Transaction ingestion and sourcing data.Week 2
  5. 05Taxability, sourcing and rate-date logic.Week 2, Week 3
  6. 06Certificate matching and coverage checks.Week 3
  7. 07Return preparation and ledger tie-out.Week 3, Week 4
  8. 08Filing-portal and remittance integration.Week 5
  9. 09Evaluation suite and regression periods.Week 4, Week 5
  10. 10Guardrails and filing-authorisation controls.Week 4
  11. 11Observability and trace 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 state and its localsProductionProduction filing cycleAdvancedMulti-state and home rule
Introduced at Pilot
Return preparation and tie-out✓✓✓
Human authorisation before filing✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Client taxability positions—✓✓
Exemption-certificate checks—✓✓
Authorised filing packages—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Home-rule and local schedules——✓
Nexus and calendar monitoring——✓
Multi-entity, high volume——✓
Enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on jurisdictions in scope, transaction volume, certificate coverage, filing 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 registrations, filing calendars and portal access

    Registration footprint and filing-calendar mapping

    Week 1
  2. 02Who may submit a return and release the remittance

    Filing-authority definition and authorisation controls

    Week 1
  3. 03Access to the ERP, billing and marketplace transaction data

    Source-system and tax-engine assessment, then ingestion

    Week 2
  4. 04Your agreed taxability positions by product and service

    Taxability, sourcing and rate effective-date logic

    Week 3
  5. 05Exemption certificates and the customers they cover

    Certificate matching, coverage checks and expiry handling

    Week 3
  6. 06Recent returns, notices and any amended filings

    Evaluation suite, regression periods and failure-mode testing

    Week 4
  7. 07Named reviewers or test users

    Authorisation workflow, then the pilot filing cycle and 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. Week 5 runs evaluation and the first supervised filing cycle together.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Workflow discovery, filing authority and registration footprint

W2Transaction ingestion, engine access and the determination baseline

W3Taxability and sourcing logic, certificates and the ledger tie-out

W4Evaluation suite, authorisation guardrails and failure-mode testing

W5Portal and remittance integration, then a supervised filing cycle

W6One live period filed under authorisation, then Agent Care handover

Reading the band
Bars cover only the weeks their work is named in. Nothing is filed before week 5, and only under authorisation.
At the end of W6
One live period is prepared, tied out, authorised and filed by your team, then Agent Care monitors.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Accounting AI agent

Build a sales-tax filing agent around your compliance calendar.

Show us your registrations, transaction sources and who signs the returns. We'll work back from the authorisation point, agree what can be prepared unattended and scope a pilot on one filing period.

Typical delivery
4–6 weeks
Deployment
Your indirect-tax stack
Human review
Before filing
After launch
Agent Care available

Nestack Agents · Sales-tax & compliance filing

AGT-ACC-07 · Agent Care available after launch

Build a sales-tax filing agent

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

About you
Your books

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