Main site

Accounting AI agent · Financial close

Month-End-Close Orchestration AI Agent

Run the close calendar, check dependencies and gates, chase owners, assemble evidence and report status — with cut-off rules, evaluation, audit trails and a named human signing every task.

Typical delivery
4–6 weeks
Deployment
Your stack
Human review
Every sign-off
After launch
Agent Care

What this agent does

Automates the coordination layer of the close

In

  1. 01

    Ingest the close calendar, task list, owners and dependency order from your close or task system.

  2. 02

    Read period status, sub-ledger locks, reconciliation state and prepared journals from the ledger.

Reason

  1. 03

    Test each task's dependencies and cut-off rules against the period state actually recorded.

  2. 04

    Check that the evidence a gate requires — reconciliation, support, reviewer sign-off — is attached and current.

  3. 05

    Apply the client's own calendar, gate definitions and materiality thresholds, not a generic checklist.

Decide

  1. 06

    Detect blocked, late, out-of-sequence and unevidenced tasks, and intercompany or revaluation gates still unmet.

  2. 07

    Route each exception to its named owner or reviewer with what is missing and what it blocks.

Out

  1. 08

    Retain the readiness check, evidence links, chase history and reviewer decisions for every task.

  2. 09

    Prepare schedules, status and the close pack for approval — closing the period stays a human action.

Product statement

The agent coordinates, checks and prepares; closing a period, posting material journals and signing off a task stay with a named human.

Example workflow

One close task, end to end

  1. 1Task becomes dueClose calendar, task list, named owner and the tasks it sits behind
  2. 2Period state gatheredSub-ledger locks, reconciliation status, open items, prior-period reversals and prepared journals
  3. 3Readiness proposedDependencies met, cut-off respected, evidence attached and confidence
  4. 4Gate checks appliedReconciliation signed, intercompany agreed, revaluation complete, variance explained and thresholds applied

No human action required

Stages 1 to 4 run without a person in the loop — a task whose dependencies are met and whose evidence is attached reaches the gate unaided.

5DecisionSplits on the confidence threshold
High confidence

Dependencies met and evidence attached — follows the approved path.

Low confidence

Blocked, out of sequence or unevidenced — enters human review.

Human review

The task is held with its dependency check, the evidence it is missing and confidence.

Approve · Correct · Request review

Approved — handed back

  1. 6Status updated and owner chasedOnly where write access and approval policy allow it; the period is not closed
  2. 7Outcome evaluatedGate breaches caught, first-pass completions, chases needed, overrides and post-close adjustments

Overrides

Reviewer decisions on gates and sequence are counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Closing or re-opening an accounting period.
    • Signing off a close task or reconciliation.
    • Posting a journal above the materiality threshold.
    • Approving a top-side or consolidation adjustment.

  • Automation boundary

    Agent acts unaided

    • ✓Assemble each task's dependency, cut-off and evidence state.
    • ✓Test gate conditions against the period and sub-ledger records.
    • ✓Chase named owners and reviewers for what is still missing.
    • ✓Route exceptions to review and retain the readiness evidence.

    Write actions run only inside the approval boundaries agreed during implementation. Closing the period is not one of them.

    • Waiving a gate to keep the close moving.
    • Accepting an intercompany difference as agreed.
    • Publishing the management pack or reported figures.
  • Changing the close calendar, gates or thresholds.

Example output

One close task, annotated

Everything the agent reports is attached to the task it came from.

Readiness output · single close task

Illustrative example
Close task
Accruals reconciliation
Depends on
AP sub-ledger lock
Difference
$3,200.00
Readiness
Held — gate unmet
Confidence
93%
Blocks
Flux review, reporting pack

As received

Taken from the close calendar and the period record in the ledger — nothing on this side is inferred.

Evidence checked

  • AP sub-ledger still open
  • Reconciliation unsigned
  • Difference over threshold

Why it is held

Dependency and sign-off are both unmet — the agent reports the gate, it cannot waive it.

What the score decides

Below the configured threshold the task routes to review. The sign-off itself is a human action either way.

Action

ApproveCorrectRequest 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 close task

From the close calendar

03Readiness

Apply client-specific context

Use the client's own close calendar, dependency order, gate definitions and thresholds.

01Approved path

Reduce routine chasing and status work

Tasks with their dependencies met and their evidence attached are checked and reported without manual follow-up.

02Human review

Focus the close team on blockers

Blocked, out-of-sequence and unevidenced tasks move to review instead of every task being chased by hand.

04

Build an evidence trail

Retain the readiness check, evidence links, chase history, confidence, evaluator result and human 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.

  • ERP & general ledger

    • Oracle
    • SAP S/4HANA
    • Workday Financials
    • NetSuite
    • Sage Intacct

  • Sub-ledgers

    • AP
    • AR
    • Payroll
    • Fixed assets
    • Revenue

  • Close & reconciliation

    • Close-management tools
    • task trackers
    • Reconciliation platforms

Agent

Month-end-close orchestration

  • Reads close state
  • Checks gates
  • Chases owners

  • Reporting & chasing

    • Consolidation
    • reporting packs
    • Email
    • Slack/Teams
    • ticketing

  • 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 your sign-off

Each control wraps the one inside it. A task clears every layer before its status changes, and closing the period sits outside all six.

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 model, prompt, calendar and gate-rule changes.Track
L4TraceabilityRecord state read, gate result, chase, action and override.Record
L3Human sign-offDefine which tasks the agent may mark and who signs.Gate
L2Dependency gatesHold a task until its inputs and period locks clear.Block
L1Confidence thresholdsLow-confidence readiness calls require review.Require review
Model coreReadiness proposed — dependencies, gate result, evidence and confidence
L1 – L2
Decide whether the task may advance
L3
Decides who signs and who closes the period
L4 – L5
Keep the evidence and the calendar honest
L6
Pulls automation back when signals degrade

How Nestack evaluates it

Evaluate the full workflow — not only the final status.

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

Surface — the close status the controller sees

  1. E1Final-output evaluationWas the readiness call for that task correct?
  2. E2Step-level evaluationDid the agent check the right dependencies, gates and thresholds?
  3. E3Tool evaluationDid it read the correct period, sub-ledger and reconciliation?
  4. E4Confidence calibrationDo low-confidence readiness calls actually contain more errors?
  5. E5Slice evaluationHow does performance change across specific close cohorts?
  6. E6Business outcomeDid the close land on time, and what was adjusted after it?

Floor — the period a named human signs off

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

    • MC-01

      Evidence not retrieved

      Support sits in a thread or folder the agent cannot read.

    Stage gathers

    Close calendar, task state, period locks and evidence

  2. 02 · Sequencing

    2 modes

    • MC-02

      Dependency violation

      A task is released before the input it depends on closes.

    • MC-03

      Cut-off breach

      A late item is scheduled into a period already locked.

    Stage proposes

    Readiness, gate result, owner and confidence

  3. 03 · Tool / write

    2 modes

    • MC-04

      Duplicated accrual reversal

      A reversal is queued twice, or the prior one never ran.

    • MC-05

      Intercompany left unbalanced

      A residual is recorded as agreed and carried forward.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • MC-06

      Green over an unmet gate

      Status reports complete on a task with no evidence attached.

    Stage returns

    The close status the controller and reviewer see

  5. 05 · Change / Version

    1 mode

    • MC-07

      Reopening not propagated

      Downstream sign-offs survive a reopening that voided them.

    Stage tracks

    Model, prompt, calendar and gate-rule changes

  • Sev-1 · close advances past an unmet gate
  • Sev-2 · wrong figure carries into reporting
  • Sev-3 · evidence missing, task returns to owner
See the complete Accounting failure-mode catalogue

Affected slices

A close that lands on time can hide concentrated risk

Aggregate close performance can look acceptable while a small number of task cohorts carry most of the gate breaches and nearly all of the adjustments booked after sign-off. Nestack reports performance by slice, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Multi-entity intercompany tasks5.1%3.3×Review
Foreign-currency revaluation tasks3.9%2.5×Review
Re-opened and adjusted periods2.9%1.9×Watch
Steady-state single-entity tasks1.1%0.7×Normal

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

2 of 4 slices over threshold

Evidence-linked improvement

Each breach becomes a standing gate check

The close repeats every month, so each cycle is a dated test of the last one's fix — a breach found in one period becomes a check the next one has to pass.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Gate breaches, late tasks or post-close adjustments rise in a cohort.

  2. 02

    Diagnose

    Traced to a stale dependency, a gate rule, an evidence source or the owner routing.

  3. 03

    Improve

    The dependency, gate rule or threshold is re-approved by the controller and version-linked.

  4. 04

    Verify

    The change is re-run against the recorded state of the close that failed.

  5. 05

    Learn

    That breach becomes a standing check and the calendar change enters the runbook.

  6. Learn → Detect

    The return edge. Next month's close is measured against every gate the last one failed.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, calendar and gates, agent workflow, evaluation, then a shadow close and handover.

Workstream

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

  1. 01Close-calendar and dependency-map discovery.Week 1
  2. 02ERP, sub-ledger and close-tool API assessment.Week 1, Week 2
  3. 03Gate, evidence and threshold definition.Week 1, Week 2
  4. 04Task, owner and period-state ingestion.Week 2
  5. 05Dependency and cut-off checking logic.Week 2, Week 3
  6. 06Confidence scoring and exception routing.Week 3
  7. 07Chase, escalation and notification workflow.Week 3, Week 4
  8. 08Human review and sign-off workflow.Week 4
  9. 09Guardrails, period-lock and approval controls.Week 4
  10. 10Evaluation suite and regression closes.Week 4, Week 5
  11. 11Status reporting and close-pack preparation.Week 5
  12. 12Observability, deployment and Agent Care handover.Week 5, 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 closeProductionProduction integrationAdvancedMultiple entities / systems
Introduced at Pilot
Readiness and gate checks✓✓✓
Human sign-off✓✓✓
Baseline evaluation✓✓✓
Introduced at Production
Dependency and cut-off logic—✓✓
Evidence collection and linking—✓✓
Owner chasing and escalation—✓✓
Approved status write-back—✓✓
Observability and evaluation—✓✓
Introduced at Advanced
Intercompany and revaluation gates——✓
Multi-entity consolidation close——✓
Enterprise controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on entity count, ERP and close-tool integrations, calendar and gate complexity, 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 close calendar, task list and named owners

    Close-calendar and dependency-map discovery

    Week 1

  2. 02Gate definitions, evidence requirements and thresholds

    Gate, evidence and threshold definition

    Week 1

  3. 03Access to the ERP, sub-ledgers and close tooling

    API assessment, then task, owner and period-state ingestion

    Week 2

  4. 04The dependency order between tasks and sub-ledgers

    Dependency and cut-off checking logic

    Week 2

  5. 05Materiality thresholds and period-lock policy

    Guardrails, period-lock and approval controls

    Week 4

  6. 06Recent closes, including one that ran late

    Evaluation suite, regression closes and failure-mode testing

    Week 4

  7. 07Named preparers, reviewers and the controller

    Sign-off workflow, then a shadow close 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. The close runs once a month, so week 5 is timed to land on a real one.

Phase

W1

W2

W3

W4

W5

W6

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

Week focus

W1Close-calendar mapping, dependency order and the automation boundary

W2ERP, sub-ledger and close-tool ingestion, and the readiness baseline

W3Gate checks, confidence logic and the chase and escalation workflow

W4Evaluation suite, period-lock guardrails and failure-mode testing

W5Shadow run against a live close, reporting status but writing nothing

W6Approved write-back, production validation and Agent Care handover

Reading the band
Bars span only the weeks their work is named in. Evaluation and the shadow close genuinely overlap in week 5.
At the end of W6
One month-end close has run in shadow and been checked task by task, then Agent Care takes over monitoring.
Duration
Six-week plan shown · typical delivery 4–6 weeks, sequenced so the shadow run lands on a live close.

Next step · Accounting AI agent

Build a close orchestration agent around your close calendar.

Show us your close calendar, the dependency order between tasks and the gate each one has to clear. We'll walk the calendar task by task, agree which gates the agent may check and start where the close stalls.

Typical delivery
4–6 weeks
Deployment
Your finance stack
Human review
Every task sign-off
After launch
Agent Care available

Nestack Agents · Month-end-close orchestration

AGT-ACC-06 · Agent Care available after launch

Build a close orchestration 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.