S SARAH LOCKETT

04 / AI case study / Multi-entity accounting

Designed · not yet built

The Month-End Close Copilot

A close checklist that runs itself: task states tracked, workbook hand-offs validated, and variance commentary drafted for the human who signs off.

TYPEFinance operations
PLANNED BUILDPlanned · ~3 weeks
STATUSDesigned · not yet built

PROVEN EXPERIENCE → NEW BUILD

Rooted in a close I ran by hand

After a leadership transition at EdgeWater, I stepped in and ran month-end close for two entities through a twelve-workbook Excel chain built on Power Query, pivot tables, and nested logic, and reported results to Sweden. This case study is that close, rebuilt so it cannot be lost with its operator.

THE BRIEF

An orchestration layer over a multi-workbook close: the copilot tracks checklist state, validates that each workbook's outputs tie to the next one's inputs, drafts variance commentary against prior periods, and assembles the review packet.

The problem

A month-end close that lives in a chain of interdependent workbooks depends on one person knowing the order, the hand-offs, and what a bad number looks like. When that person leaves, the close becomes archaeology.

The system

The copilot holds the checklist and the dependency map, validates tie-outs at every hand-off (totals that must agree, balances that must roll), drafts the variance commentary with linked evidence for human review, and assembles the close packet. The accountant decides; the copilot remembers, checks, and drafts.

  1. Close checklist
  2. Dependency map
  3. Tie-out validation
  4. Variance analysis
  5. Commentary drafts
  6. Review packet
ExcelPower QueryClaude agentsPower AutomateSynthetic close chain

DESIGNED OUTCOMES

What “working” will mean.

  • A close that survives its operator
  • Tie-out failures caught at the hand-off, not at review
  • Variance commentary drafted with evidence attached
  • A review packet assembled, not compiled by hand

MEASUREMENT PLAN

How it gets judged.

  • Tie-out break detection vs. seeded truth
  • Commentary accuracy
  • Close cycle time
  • Checklist completion visibility
  • Reviewer edit distance

THE BUILD PLAN

How this gets built
when it gets built.

Days 1–5

Model

Build a synthetic multi-workbook close chain with seeded tie-out breaks and variance stories.

Days 6–11

Validate

Dependency tracking and automated tie-out checks across the chain.

Days 12–16

Narrate

Variance detection and commentary drafting against prior periods, evidence linked.

Days 17–21

Package

Review-packet assembly, seeded-truth evaluation, and the written case study.

Transparency note. This is a self-directed AI case study built on synthetic data. Results will be reported as evaluation findings against seeded ground truth, not client outcomes. When the build is complete, this page gains: working close orchestration, tie-out validation report, synthetic close-chain workbooks, drafted variance commentary, close runbook, written case study.

NEXT CASE STUDY / 01The Workflow X-Ray
→