Main site

Accounting AI agent · Audit preparation

Audit-Preparation Copilot (Preparer-Side)

Assemble the audit pack — lead schedules tied to the trial balance, evidence matched to each request and prior-year figures re-checked — held for preparer sign-off, with evaluation, audit trails and guardrails built in.

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

What this agent does

Assembles the pack and shows its support

In

  1. 01

    Ingest the request list, the trial balance and prior-year workpapers from supported systems.

  2. 02

    Normalise each request into entity, period, account grouping and the evidence it calls for.

Reason

  1. 03

    Build the lead schedule from the ledger and agree it to the trial balance it was drawn from.

  2. 04

    Match supporting documents to the request, the account and the period they belong to.

  3. 05

    Roll prior-year workpapers forward and re-check every carried figure against this year's data.

Decide

  1. 06

    Flag breaks in the tie-out, unsupported balances and populations that cannot be shown complete.

  2. 07

    Route estimate, judgement and disclosure items to the named preparer for sign-off.

Out

  1. 08

    Retain the request, the source extract, the schedule, the evidence and the preparer's decision.

  2. 09

    Release items to the auditor only inside the approval boundaries agreed during implementation.

Product statement

The agent assembles and cross-checks inside the approval boundaries agreed during implementation; it gives no assurance, and a named preparer signs the pack.

Example workflow

One request, end to end

  1. 1Request receivedRequest list, auditor portal or the finance team's own tracker
  2. 2Scope resolvedEntity, period, account grouping, materiality and last year's version of the same item
  3. 3Item assembledLead schedule built from the ledger, prior year rolled forward, supporting documents attached
  4. 4Controls appliedTie-out to the trial balance, entity and period checks, evidence-to-balance checks and confidence threshold

No human action required

Stages 1 to 4 run without a person in the loop — nothing has left the finance team yet, so the lane stays empty until the gate.

5DecisionSplits on the confidence and item-class gate
High confidence

Joins the pack as a tied, evidenced draft.

Low confidence

Held with the check it could not clear.

Preparer sign-off

The item is held with its schedule, source extract, evidence and the break that stopped it.

Approve · Amend · Send back

Signed — handed back

  1. 6Item enters the packSigned by the preparer before it reaches the auditor; the agent concludes nothing
  2. 7Outcome evaluatedTie-out accuracy, population completeness, preparer amendments, auditor queries and business outcome

Amendments

Corrections and send-backs made at sign-off are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Anything released to the auditor.
    • Judgements, estimates and the basis for them.
    • Disclosure wording and the disclosure checklist.
    • Written answers to auditor queries.

  • Automation boundary

    Agent acts unaided

    • ✓Build lead schedules and agree them to the trial balance.
    • ✓Match evidence to the request, entity and period.
    • ✓Roll prior-year workpapers forward and re-check every carried figure.
    • ✓Flag breaks and route the item to the named preparer.

    Items reach the auditor only inside the approval boundaries agreed during implementation.

    • Any assertion that a population is complete.
    • Re-issuing a pack after a late journal.
    • Work the audit team will later test.
  • Changes to account mappings or tie-out tolerances.

Example output

One request, annotated

Everything the agent assembles is attached to the request and the extract it came from.

Assembled item · single request

Illustrative example
Request
Accruals lead schedule
Entity
Operating subsidiary
Period
FY25
Tie-out
Agrees to the trial balance
Confidence
91%
Open break
An accrual without support

As received

The request, entity and period exactly as the request list states them — the agent does not widen the period or pull a second entity in.

Evidence used

  • Trial-balance extract
  • Prior-year signed schedule
  • Posted supplier invoices

Why it holds

Every other line agrees to posted support; one accrual rests on an estimate the ledger does not evidence, so the item does not go out whole.

What the score decides

Below the configured threshold the item is held for the named preparer instead of entering the pack.

Action

ApproveAmendSend back

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 PBC requests

From the auditor's request list

03Assembly

Apply your own ledger and close

Use the trial balance, the account groupings and last year's signed workpapers as they stand today.

01Approved path

Cut the schedule-rebuilding time

Routine, in-scope requests come back assembled, tied and evidenced instead of being rebuilt each year.

02Human review

Put preparers on the judgement items

Estimates, disclosures and unsupported balances reach a named preparer instead of every request doing so.

04

Build an evidence trail

Retain the request, source extract, assembled schedule, confidence, evaluator result and preparer 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.

  • Workpaper platforms

    • Caseware
    • CCH Engagement
    • TeamMate+
    • workpaper APIs

  • Request portals

    • Suralink
    • AuditBoard
    • Client portals
    • request-list exports

  • ERP, GL & close

    • NetSuite
    • Sage Intacct
    • FloQast
    • BlackLine
    • trial-balance exports

Agent

Audit-preparation copilot

  • Reads the ledger
  • Assembles and ties
  • Routes to sign-off

  • Documents

    • SharePoint
    • iManage
    • Document stores
    • shared drives

  • 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 audit pack

Each control wraps the one inside it. An item clears every layer before it can enter the pack.

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 ledger re-extracts, model, prompt and configuration changes.Track
L4TraceabilityRecord request, extract, schedule, evaluation, decision and amendment.Record
L3Preparer sign-offDefine which items may enter the pack without a named preparer signing.Gate
L2Assurance boundaryThe agent assembles and cross-checks; it never tests, concludes or opines.Restrict
L1Tie-out checksEach schedule must agree to the trial balance for that entity and period.Reconcile
Model coreItem assembled — schedule, tie-out, evidence, open breaks and confidence
L1 – L2
Decide whether the item may stand
L3
Decides whether a named preparer signs it
L4 – L5
Keep the record and the extract current
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the full workflow — not only the finished schedule.

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

Surface — the pack the preparer signs

  1. E1Final-output evaluationDid the assembled item answer the request as it was raised?
  2. E2Tie-out evaluationDoes every schedule still agree to the trial balance?
  3. E3Population evaluationWas the population complete, and can that be shown?
  4. E4Hold calibrationAre unsupported and judgement items actually held back?
  5. E5Slice evaluationHow does performance change across specific request cohorts?
  6. E6Business outcomeHow many auditor queries and reworked schedules followed?

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

    2 modes

    • AU-01

      Evidence coverage gap

      The document store holds nothing that supports the request as raised.

    • AU-02

      Wrong entity or period

      Support is gathered for the right account in the wrong company or year.

    Stage gathers

    Trial balance, ledger detail, prior-year files and evidence

  2. 02 · Reasoning

    2 modes

    • AU-03

      Prior year carried unchecked

      Last year's figures and groupings roll forward without being re-verified.

    • AU-04

      Population shown as complete

      A filtered or truncated extract is presented as the whole population.

    Stage proposes

    The schedule, the tie-out, the evidence and confidence

  3. 03 · Tool / write

    1 mode

    • AU-05

      Release before sign-off

      An item reaches the auditor as though a preparer had signed it.

    Stage writes

    Only where pack access and approval boundaries allow it

  4. 04 · Output

    1 mode

    • AU-06

      Evidence-to-item mismatch

      A document is filed against a balance it does not actually evidence.

    Stage returns

    The item the preparer and the audit pack see

  5. 05 · Change / Version

    1 mode

    • AU-07

      Silent version drift

      A late journal moves the ledger under a finished schedule.

    Stage tracks

    Ledger re-extracts, model and prompt changes

  • Sev-1 · unsupported work could reach the auditor
  • Sev-2 · wrong figure or wrong entity in the pack
  • Sev-3 · assembly degrades, item routes to sign-off
See the complete Accounting failure-mode catalogue

Affected slices

A pack that ships on time can still be thin

First-year requests and multi-entity consolidations take far more tie-out and evidence work than the routine ones, and that is where the failures land. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
First-year audit requests6.1%3.6×Review
Multi-entity consolidations4.2%2.5×Review
Late-journal re-runs3.1%1.8×Watch
Routine reconciliations1.1%0.6×Normal

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

2 of 4 slices over threshold

Evidence-linked improvement

What breaks once becomes a standing check

Every break found in one cohort is turned into a check that runs before assembly, so next period's pack starts from that check rather than from last year's file.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Tie-out breaks or unsupported items rise in a request cohort.

  2. 02

    Diagnose

    Failure isolated to the extract, the mapping, the evidence match or the roll-forward.

  3. 03

    Improve

    Mapping corrected or a pre-assembly check added — approved and version-linked.

  4. 04

    Verify

    Affected requests are re-assembled against the current trial balance.

  5. 05

    Learn

    The break becomes a standing check; the signed schedule becomes next year's baseline.

  6. Learn → Detect

    The return edge. The next cycle starts against a current extract and one more pre-assembly check.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, ledger and request mapping, assembly workflow, evaluation, integration, then production validation and handover.

Workstream

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

  1. 01Workflow discovery and assurance-boundary definition.Week 1
  2. 02Source-system and workpaper assessment.Week 1, Week 2
  3. 03Request-list, entity and account-grouping mapping.Week 1, Week 2
  4. 04Ledger extraction and schedule templates.Week 2
  5. 05Tie-out, evidence and roll-forward checks.Week 2, Week 3
  6. 06Confidence scoring and hold routing.Week 3
  7. 07Preparer sign-off workflow.Week 3, Week 4
  8. 08Workpaper and document-system integration.Week 5
  9. 09Evaluation suite and regression cases.Week 4, Week 5
  10. 10Guardrails and release 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. No tier adds testing or assurance.

Capability✓ in scope · — not at this tierPilotOne entityProductionProduction integrationAdvancedGroup and multi-entity
Introduced at Pilot
Assembled schedules and tie-outs✓✓✓
Preparer sign-off✓✓✓
Assurance-boundary controls✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Evidence matching and gap chasing—✓✓
Prior-year roll-forward checks—✓✓
Approved release actions—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Group and multi-entity packs——✓
Multi-stage sign-off——✓
Enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on entities in scope, request volume, source systems, sign-off 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. 01The request list and who owns each item

    Request-list, entity and account-grouping mapping

    Week 1

  2. 02The assurance boundary and who signs the pack

    Assurance-boundary definition and release controls

    Week 1

  3. 03Access to the ledger, trial balance and document store

    Source-system and workpaper assessment

    Week 2

  4. 04Last year's signed workpapers and lead schedules

    Schedule templates and prior-year roll-forward checks

    Week 2

  5. 05Tie-out tolerances and the thresholds for holding an item

    Confidence scoring, tie-out checks and hold routing

    Week 3

  6. 06The auditor queries that cost you most last year

    Evaluation suite, regression cases and failure-mode testing

    Week 4

  7. 07Named preparers or test users

    Preparer sign-off workflow, then pilot pack 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. Build waits on week 1 because the assurance boundary decides what the agent may assemble at all.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Workflow discovery, assurance boundary and request-list mapping

W2Ledger and document access, schedule templates and the assembly baseline

W3Assembly workflow, tie-out checks and preparer sign-off controls

W4Evaluation suite, release guardrails and failure-mode testing

W5Workpaper-platform integration, pilot pack and targeted corrections

W6A live request cycle under sign-off, verification and Agent Care handover

Reading the band
Bars cover only the weeks their work is named in. Evaluate and Pilot both sit in week 5 — the first live pack is assembled while the suite is still running.
At the end of W6
A live request cycle has been assembled and signed, then Agent Care monitors tie-outs, holds and the queries the pack still attracts.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Accounting AI agent

Build an audit-preparation copilot around your close.

Show us your request list, your trial balance and last year's workpapers, and tell us who signs the pack. We'll rebuild one item the way your preparers would, then agree what the agent may assemble unsigned.

Typical delivery
4–6 weeks
Deployment
Your close and workpaper stack
Human review
Before the pack goes out
After launch
Agent Care available

Nestack Agents · Audit-preparation copilot

AGT-ACC-05 · Agent Care available after launch

Build an audit-preparation copilot

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.