Tax-Provision AI Agent

What this agent does

Computes the provision, never the position it takes

Current tax is built from the trial balance, the tax basis and each entity's configured elections.

Deferreds move through a maintained inventory tracked to reversal, and Pillar Two runs separately.

Reason

The rate reconciliation is assembled by disclosure category, with taxes paid tracked by jurisdiction alongside it.

Apportionment, elections and jurisdiction rules apply as configured, never chosen position by position.

Every computed line is bound to the trial-balance or tax-basis value behind it, and gaps are marked.

Decide

Positions are screened against the recognition threshold, and only a position that clears it moves to measurement.

Valuation-allowance evidence, positive and negative, is assembled for the tax director's weighing.

Out

Every position and every allowance judgement routes to the named tax director for acceptance.

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

→ Product statement

The agent computes the provision and assembles the disclosure inputs; the tax director concludes on every position and allowance, and stays the signer of record.

Example workflow

One provision, computation to conclusion

AgentHuman

1Trial balance receivedGeneral ledger trial balance, tax-basis schedule or prior-period workpaper

2Deferred inventory rolled forwardTemporary differences, elections and prior positions, each tied to the entity and jurisdiction they came from

3Provision computedCurrent tax, deferred tax, rate reconciliation, flagged positions and confidence

4Controls appliedRecognition-threshold checks, allowance-evidence checks, jurisdiction rules and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing is booked at any of them — the agent is computing, and the tax director's lane opens at the confidence gate.

5DecisionBranches at the confidence threshold

High confidence

Goes to the tax director to review.

Low confidence

Adds a technical-tax read first.

Director review

The provision is held with its computation, its flagged positions and the confidence.

Approve · Edit · Send to technical review

Approved — ready to book▼

6Provision system updatedOnly where write access and approval policy allow it

7Outcome evaluatedComputation accuracy, director edits, allowance outcomes and post-close corrections

Edits

Every director edit is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Concluding an uncertain tax position.

Setting or releasing a valuation allowance.

Booking a tax entry in the ledger.

Signing the tax footnote.

Automation boundaryAgent acts unaided

✓Compute current and deferred tax from the trial balance and the tax basis.

✓Apply the entity's configured elections, apportionment.

✓Bind each computed line to the trial-balance or tax-basis value it came from.

✓Flag positions, allowance judgements and discrete.

Any computed line runs inside the boundaries agreed at implementation, never ahead of the tax director's review.

Determining an interim discrete item.

Characterising a difference as permanent or temporary.

Approving a return-to-provision true-up.

Changes to computation rules or automation thresholds.

Example output

One position in the provision, annotated

Everything the agent computes is attached to the basis it came from.

Provision output · single positionIllustrative example

Position

Computed line

Amount

Source of record

Confidence

Tax director of record

State research-credit position

Recognition threshold cleared on technical merits — measurement not yet run

$1,150,000

State credit study on file

88%

Tax director name and title

As receivedTaken from the trial balance and the credit study on file — nothing on this side is a conclusion the agent reached.

Evidence used State credit study on file Prior-period workpapers Trial-balance tie-out

Why this lineThe technical-merits test is met in the record provided; whether the benefit is realized.

ActionApproveEditSend to technical review

What the score decidesBelow the configured threshold the position picks up a technical-tax read before it.

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value landsValue 01 – 04

Every positionFrom the trial balance

03Computation

Compute from the basis

Draw on the trial balance, the tax basis and the entity's configured elections.

01Approved path

Recognition comes before

Routine current and deferred computations arrive already built, tied to the basis.

02Human review

Send review to the open positions

Uncertain positions and low-confidence allowance judgements are marked, so the tax director's read starts where risk concentrates.

04Build an evidence trail

The position, the authority it was weighed against and the tax director who accepted it stay on the provision.

Integrations

Typical integrations

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

ERP and general ledgerSAP · Oracle · NetSuite
Workday · Dynamics 365

Tax-provision softwareCorptax · ONESOURCE
Longview Tax · TaxStream

Workpaper & basis toolsBasis workpapers · roll-forwards
Fixed-asset and depreciation systems

Agent

Tax-provision computation

Reads the basis
Computes the provision
Holds for review

Disclosure and reportingXBRL and disclosure tools
Audit and review systems

Observability & evaluationOpenTelemetry · 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 filed provision

Each control encloses the one beneath. What none of them catches is named in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to schedule assembly when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt, computation-rule and configuration changes.Track

L4Provision trailRecord the trial balance, the computation, the flagged positions and the director's disposition.Record

L3Director reviewHold the provision for the named tax director; it governs release, not whether a position is right.Gate

L2Policy guardrailsTest computations against recognition-threshold and jurisdiction rules; a failure holds the item.Restrict

L1Confidence thresholdsRoute low-confidence computations to a technical-tax read before the director sees them.Require review

Model coreProvision computed — current, deferred, rate reconciliation, flagged positions and confidence

L1 – L2Test whether a line may stand

L3Puts the conclusion in the tax director's hands

L4 – L5Keep the position and the authority behind it

L6Drops to schedule assembly when signals degrade

How Nestack evaluates it

Evaluate the full provision workflow — not only the filed rate.

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

Surface — the rate the filing shows

Depth of coverage ▼

E1Final-output evaluationDid every computed line match its trial-balance or tax-basis source?

E2Step-level evaluationDid the agent use the right basis, elections and jurisdiction rules?

E3Tool evaluationDid it read and write the correct entity, period and jurisdiction?

E4Confidence calibrationDo low-confidence computations actually attract more director edits?

E5Slice evaluationHow does performance change across specific entity or position types?

E6Business outcomeHow many lines needed a director edit or a correction after close?

Floor — the outcome the tax director answers for

Failure modes

Where each failure originates in the agent

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

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

TX-03

Superseded basis snapshot

Temporary differences are computed from a trial balance the client later restated.

Stage gathersTrial balance, tax basis, prior positions and entity elections

02 · Reasoning2 modes

TX-04

Recognition-measurement collapse

A probability-weighted benefit is booked before the recognition threshold is cleared.

TX-06

Discrete item folded into rate

A one-time item is blended into the estimated annual rate instead of its own interim period.

Stage proposesCurrent tax, deferred tax and confidence

03 · Tool / write2 modes

TX-02

Valuation allowance auto-set

An allowance is set or released from a loss trigger alone, with no positive evidence weighed.

TX-05

True-up posted without review

A large return-to-provision true-up posts automatically without a root-cause read.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

TX-01

Disaggregation threshold missed

A required rate-reconciliation category or disaggregation threshold is missed before review.

Stage returnsThe provision the tax director reviews and accepts

05 · Change / Version1 mode

TX-07

Stale rate assumption

An enacted rate or law change is not reflected before the next computation runs.

Stage tracksModel, prompt, computation rules and jurisdiction config

Sev-1 · books outside the boundary Sev-2 · unsupported position reaches the provision Sev-3 · source degrades, computation goes to review

See the complete Accounting failure-mode catalogue →

Affected slices

A group rate can hide a single entity's risk

A group-level effective rate can look stable while a handful of entities or jurisdictions carry most of the adjustment risk. Nestack reports the provision-adjustment rate by slice, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Entities with valuation-allowance history	7.3%	3.8×	Review
Uncertain positions carried forward	5.5%	2.9×	Review
Interim periods with discrete items	3.4%	1.8×	Watch
Single-jurisdiction profitable entities	1.9%	0.7×	Normal

Bar: provision-adjustment rate lift vs. the single-jurisdiction baseline · scale 0–4.0× · tick at 2.0× 2 of 4 slices over threshold

Evidence-linked improvement

The loop closes on the next computation, not an explanation

A cycle closes when the misstatement is a regression case the next release must pass. That suite is what the next provision computed is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Provision-adjustment rate rises in an entity or jurisdiction slice.

02Diagnose

The interim rate that held until it didn't is traced back through the trial balance, the basis and the elections behind it, until the cause narrows to one.

03Improve

The fix is versioned, with the provisions that motivated it attached.

04Verify

Nothing releases until the affected computation cases pass again.

05Learn

The case is kept permanently, and the position memo changes with it.

Learn → DetectThe return edge. The next provision computed is checked 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, basis, provision workflow, evaluation, integration, then production validation and handover.

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

01Provision workflow discovery and boundary definition.

02ERP and tax-provision source assessment.

03Election and jurisdiction-rule mapping and rule mapping.

04Trial-balance ingestion and normalisation.

05Rollforward logic and basis binding.

06Confidence scoring and flag routing.

07Director review workflow.

08Provision-system integration.

09Recognition and measurement cases.

10Guardrails and acceptance controls.

11Provision-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 entity, one system ProductionProduction provision-system integration AdvancedMultiple entities / jurisdictions

Introduced at Pilot

Computing to your basis and rules✓✓✓

Director review✓✓✓

Computation-accuracy baseline✓✓✓

Introduced at Production

Reporting by jurisdiction—✓✓

Review workflow in your systems—✓✓

Approved write-back—✓✓

Provision-system integration—✓✓

Introduced at Advanced

Multi-entity provision rules——✓

Multi-stage director approvals——✓

High entity volume——✓

Multi-entity provision controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, entity volume, review controls and deployment requirements.

Separate from buildBuild 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 bringWe build with it

01Your trial balance and tax-basis structure → Trial-balance ingestion and basis mappingWeek 1

02Representative prior-period provisions → Computation baseline, rollforward and basis bindingWeek 2

03Your entity elections and valuation-allowance policies → Election, apportionment and boundary-definition mappingWeek 1

04Access to relevant APIs, feeds or exports → ERP and tax-provision source assessment, then integration setupWeek 2

05Positions you would not want booked → Valuation-allowance cases and failure-mode testingWeek 4

06What no provision may conclude alone → Confidence scoring, flag routing, guardrails and acceptance controlsWeek 3

07Named tax directors to review positions → Director review workflow, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Phases cover the weeks the work truly takes, which is why the fifth carries evaluation and pilot.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Provision workflow discovery, election mapping and the boundary W2Source integration and the computation baseline W3Computation workflow, confidence logic and acceptance controls W4Evaluation suite, guardrails and failure-mode testing W5Provision-system integration, pilot entities and targeted corrections W6One reporting cycle computed under the tax director, then Agent Care handover

Reading the bandA band covers the weeks its work is named in, and nothing else. The week 5 overlap is real, not padding.

At the end of W6Validation closes on a live provision and Agent Care picks up monitoring.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Accounting AI agent

Build a tax-provision agent around your company's review chain.

Show us your trial balance, your tax-basis workpapers and who reviews the provision today. Your tax director keeps every position and every judgement, and if your external auditor would also touch this tool, that boundary is set before deployment.

Discuss this agent →

Nestack Agents · Tax-provision computationAGT-ACC-20 · Agent Care available after launch

Tax-Provision AI Agent (ASC 740 · Current, Deferred & Rate Reconciliation)