Main site

Accounting AI agent · Tax-return prep

Tax-Return Preparation & Review AI Agent

Ingest source documents, roll the prior-year file forward, populate the return in your prep software and raise review notes with the source behind each entry — the signing preparer decides, signs and stays accountable.

Typical delivery
4–6 weeks
Deployment
Your stack
Preparer review
Pre-signature
After launch
Agent Care

What this agent does

Populates the return, never the position it takes

In

  1. 01

    A W-2, 1099, K-1 or trial balance arrives from the client portal, and the agent reads it into the return.

  2. 02

    The prior-year file rolls forward, and every carried figure keeps the source return it came from.

Reason

  1. 03

    The trial balance and source documents populate the return, form by form, in the firm's prep software.

  2. 04

    The firm's configured entity elections, apportionment rules and mapping conventions are applied, not chosen.

  3. 05

    Every populated figure is bound to the document behind it, and what the documents leave unclear is marked.

Decide

  1. 06

    A late K-1, an unreconciled 1099 or a refundable-credit claim without its documentation is flagged.

  2. 07

    Each review note reaches the signing preparer with its source document attached.

Out

  1. 08

    The source document, the populated entry, the review note and the preparer's disposition are retained against the return.

  2. 09

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

Product statement

The agent populates the return and drafts review notes; the signing preparer decides every position, signs and stays the preparer of record.

Example workflow

One return, document to sign-off

  1. 1Source documents receivedW-2, 1099, K-1, trial balance or prior-year file
  2. 2Prior year rolled forwardCarried balances, elections, entities and schedules, each tied to the return it came from
  3. 3Return populatedForm-by-form entries, review notes, flagged items and confidence
  4. 4Controls appliedSource-document checks, due-diligence checklist, apportionment rules and confidence threshold

No human action required

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

5DecisionBranches at the confidence threshold
High confidence

Goes to the signing preparer to review.

Low confidence

Adds a due-diligence review first.

Preparer review

The return is held with its source documents, its review notes and the confidence.

Approve · Edit · Send to reviewer

Approved — ready to sign

  1. 6Prep software updatedOnly where write access and approval policy allow it
  2. 7Outcome evaluatedEntry accuracy, preparer edits, due-diligence outcomes and post-review corrections

Edits

Every preparer edit is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Signing the return.
    • Transmitting before Form 8879 is signed.
    • Deciding whether a position has substantial authority.
    • Determining filing status without review.

  • Automation boundary

    Agent acts unaided

    • ✓Populate the return from source documents and.
    • ✓Apply the firm's configured entity, election and apportionment rules.
    • ✓Bind each entry to the document it came from, and mark what is unclear.
    • ✓Flag due-diligence gaps and unusual items, and hold.

    Any populated field runs inside the boundaries agreed at implementation, never ahead of a signed consent or a review.

    • Deciding whether to disclose a position on Form 8275.
    • Sending taxpayer data to a third party without a signed §7216 consent.
    • Attesting to Form 8867 due diligence.
  • Changes to mapping rules or automation thresholds.

Example output

One entry in the return, annotated

Everything the agent populates is attached to the document it came from.

Preparation output · single return

Illustrative example
Return
Partnership K-1 return
Populated entry
Guaranteed payment carried to Schedule E as ordinary income
Amount
$42,000
Source document
K-1, Box 4a
Confidence
91%
Preparer of record
Preparer name and PTIN

As received

Taken from the K-1 and the prior-year file — nothing on this side is a position the agent decided.

Source documents used

  • K-1, Box 4a
  • Partnership agreement
  • Prior-year Schedule E

Why this entry

The K-1 states it as a guaranteed payment; whether it changes.

What the score decides

Below the configured threshold the return picks up a due-diligence read before it reaches.

Action

ApproveEditSend to reviewer

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 return

From the source documents

03Preparation

Populate from the documents

Draw on the documents on file and the firm's configured elections and apportionment rules.

01Approved path

The signature is the position

Routine entries arrive already populated, tied to their documents.

02Human review

Send review to the risky returns

Due-diligence gaps and low-confidence entries are marked, so the preparer's read starts where risk concentrates.

04

Build an evidence trail

The entry, the document it came from and the preparer who signed stay on the return.

Integrations

Typical integrations

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

  • Tax prep software

    • Drake
    • UltraTax
    • Lacerte
    • ProSeries
    • CCH Axcess

  • Source-document intake

    • SafeSend
    • Intuit Link
    • K-1 aggregation portals

  • Practice management

    • Client portals
    • e-organizers
    • Prior-year engagement files

Agent

Tax-return preparation & review

  • Reads the documents
  • Populates the return
  • Holds for review

  • Signature and transmission

    • Form 8879 e-signature
    • ERO / e-file 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 filed return

Every control wraps the one within it. What the set does not catch is named in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to data capture when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt, mapping-rule and configuration changes.Track
L4Return trailRecord the source document, the entry, the review note and the preparer's disposition.Record
L3Preparer reviewHold the return for the signing preparer; it governs release, not whether a position is right.Gate
L2Diligence guardrailsTest entries against the due-diligence checklist and consent status; a failure holds the return.Restrict
L1Confidence thresholdsRoute low-confidence entries to a due-diligence read before the preparer sees them.Require review
Model coreReturn populated — entries, review notes, flagged items and confidence
L1 – L2
Test whether an entry may stand
L3
Puts the signature in the preparer's hands
L4 – L5
Keep the entry and the document behind it
L6
Narrows to data capture when signals degrade

How Nestack evaluates it

Evaluate the full preparation workflow — not only the filed position.

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

Surface — the return the client signs

  1. E1Final-output evaluationDid every populated entry match its source document?
  2. E2Step-level evaluationDid the agent use the right prior-year file, elections and apportionment rules?
  3. E3Tool evaluationDid it read and write the correct return and the correct field?
  4. E4Confidence calibrationDo low-confidence entries actually attract more preparer edits?
  5. E5Slice evaluationHow does performance change across specific return types?
  6. E6Business outcomeHow many entries needed a preparer edit or a correction after review?

Floor — the outcome the firm signs for

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

    • RP-03

      Superseded prior-year source

      A rolled-forward balance is read from a year the client later amended.

    Stage gathers

    Source documents, prior-year file and firm rules

  2. 02 · Reasoning

    2 modes

    • RP-04

      Unsupported position drafted

      An entry is drafted with no citation the substantial-authority standard would accept.

    • RP-06

      Credit claimed without diligence

      A refundable credit is populated from rolled-forward data with no current-year interview on file.

    Stage proposes

    Populated entries, flags and confidence

  3. 03 · Tool / write

    2 modes

    • RP-02

      Unconsented data transfer

      Source documents reach a third-party process before a signed §7216 consent is on file.

    • RP-05

      Stale election carried forward

      A prior-year method or election rolls into this year's return without a reviewer prompt.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • RP-01

      Preparer identifier omitted

      The signing preparer's PTIN is missing from the populated return before it reaches review.

    Stage returns

    The return the signing preparer reviews and signs

  5. 05 · Change / Version

    1 mode

    • RP-07

      Silent mapping regression

      A software or rule-set update widens what the agent will populate without a reviewer prompt.

    Stage tracks

    Model, prompt, mapping rules and apportionment config

  • Sev-1 · transmits outside the boundary
  • Sev-2 · unsupported entry reaches the return
  • Sev-3 · source degrades, entry routes to review
See the complete Accounting failure-mode catalogue

Affected slices

One review-note rate can hide where the risk sits

A review sample drawn evenly across return types can look acceptable in aggregate while a handful of cohorts absorb most of the review notes. Nestack reports the review-note rate by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Partnership K-1s received late6.9%3.6×Review
Multi-state apportionment returns5.1%2.7×Review
Returns with refundable credits3.3%1.7×Watch
Single-state wage-income returns1.8%0.7×Normal

Bar: review-note-rate lift vs. the single-state wage-income 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, not an explanation

The loop shuts when the wrong position is a regression case, not when it has been explained. That suite is what the next return prepared is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Review-note rate rises in a return-type slice.

  2. 02

    Diagnose

    The reviewer who accepted it retraces the entry to its document until the cause narrows to one.

  3. 03

    Improve

    Whatever changes ships against a version, with the returns that prompted it attached.

  4. 04

    Verify

    Nothing ships until the affected positions pass a second time.

  5. 05

    Learn

    The suite grows by one case; so does the review checklist.

  6. Learn → Detect

    The return edge. The next return prepared 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, documents, preparation workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and boundary definition and it is logged..Week 1
  2. 02Document and software source assessment.Week 1, Week 2
  3. 03Election and apportionment-rule mapping and rule mapping.Week 1, Week 2
  4. 04Document ingestion and normalisation.Week 2
  5. 05Roll-forward logic and document binding.Week 2, Week 3
  6. 06Confidence scoring and review-note routing.Week 3
  7. 07Preparer review workflow.Week 3, Week 4
  8. 08Prep-software and e-file integration.Week 5
  9. 09Position and due-diligence cases.Week 4, Week 5
  10. 10Guardrails and review controls.Week 4
  11. 11Return-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 prep-software seatProductionProduction prep-software integrationAdvancedMultiple offices / entities
Introduced at Pilot
Populating to your documents and rules✓✓✓
Preparer review✓✓✓
Preparation-accuracy baseline✓✓✓
Introduced at Production
Reporting by return type—✓✓
Review workflow in your systems—✓✓
Approved write-back—✓✓
Prep-software integration—✓✓
Introduced at Advanced
Multi-jurisdiction rules——✓
Multi-stage preparer approvals——✓
High return volume——✓
Multi-state return controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, return volume, review 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 prep-software chart and entity structure

    Document ingestion and entry mapping

    Week 1

  2. 02Representative prior-year returns

    Roll-forward baseline and entry-to-document binding

    Week 2

  3. 03Your firm's elections and due-diligence policies

    Election, apportionment and boundary-definition mapping

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Prep-software and document-source assessment, then integration setup

    Week 2

  5. 05Positions you would not want signed

    Credit cases and failure-mode testing

    Week 4

  6. 06What no return may assert unreviewed

    Confidence scoring, flag routing, guardrails and review controls

    Week 3

  7. 07Named signing preparers to review drafts

    Preparer review 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

Bands follow the real work rather than the plan, which is why evaluation and pilot share week 5.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Return workflow discovery, election mapping and the automation boundary

W2Source integration and the roll-forward baseline

W3Preparation workflow, confidence logic and review controls

W4Position and due-diligence cases, guardrails and failure-mode testing

W5E-file integration, pilot returns and targeted corrections

W6One filing cycle prepared under the signing preparer, then handover

Reading the band
A band covers the weeks its work is named in, and no others. The week 5 overlap is real, not padding.
At the end of W6
Once the cycle validates, Agent Care owns the running agent.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Accounting AI agent

Build a return-preparation agent around your firm's review chain.

Show us your source documents, your prep software and who signs. The position, the signature and the transmission stay with your preparer of record.

Typical delivery
4–6 weeks
Deployment
Your prep software
Preparer review
Pre-signature
After launch
Agent Care available

Nestack Agents · Tax-return preparation & review

AGT-ACC-16 · Agent Care available after launch

Build a return-preparation agent

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

About you
Your returns

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