ع
Start Topics Teams Reference What's new Saved
playbook

Run month-end close, end to end

Clean the export, reconcile the accounts, roll up totals with shown work, compare to last month, model the runway, and write the leadership summary — a repeatable close where every number is verified and defensible.

advanced ~half a day
when to reach for this

Month-end close is where every Finance skill gets tested at once, under a deadline. Done ad hoc, it's a frantic afternoon of half-trusted spreadsheets ending in a summary you're not sure you can defend. This system walks the whole arc — clean the file, reconcile the accounts, total with shown work, compare to last month, check the runway, and write the briefing — so close is a checklist you run, not a panic you survive. It ties the other five Finance playbooks into one defensible workflow where every number traces back to a verified step.

gather this first
  • The raw exports for the month — billing-export-may.csv, payments-received.csv, bank-statement.csv — exactly as they came out, plus last month's clean versions to compare against.
  • Your standing artifacts: recurring-costs.csv, your categorization-rules.md (so every total uses the same buckets as last month), and your saved cashflow-model.md and leadership-summary-may.md templates from the earlier playbooks.
  • The deadline and the audience: when the close is due, and who reads the summary — so you know how tight and how defensible it needs to be.
the workflow

  1. Clean and profile every input first

    Nothing downstream can be trusted if the inputs aren't. In Claude Desktop, open the folder with the month's raw exports and ask in the chat — no terminal needed. Run the Make a messy export trustworthy loop on each file before anything else — one dirty export poisons the whole close.

    you ask
    We're running month-end close. Start by profiling each file — billing-export-may.csv, payments-received.csv, bank-statement.csv: row count, columns, date range, total, and any blank rows, duplicate IDs, or mismatched dates. Propose fixes for me to approve, then save clean versions. Don't move to reconciliation until each file is clean and re-verified.

    what you get back A clean, re-verified version of each file with a before/after total per file ("billing: 1,248 → 1,244 rows, total unchanged at 117,950"). You start the close knowing every input is trustworthy and why the row counts moved.

    This is Make a messy export trustworthy run three times. Resist the urge to skip it on the 'clean-looking' file — that's the one that bites you.

  2. Reconcile the accounts

    With clean inputs, run the Reconcile two files, line by line system: match payments to invoices and ledger to bank, surface every exception, and make the totals tie out before you total anything.

    you ask
    Now reconcile: match payments-received against the invoices by invoice number, and the ledger against bank-statement by transaction reference. Give me the matched/unpaid/orphan buckets with totals for each, the outstanding total, and confirm each reconciliation ties out. Flag any orphan payment to investigate before we close.

    what you get back Reconciliations that balance — "Invoiced 64,600 = Paid 58,200 + Unpaid 6,400; bank ties to ledger with 1 orphan of 600 to investigate" — and a short exception list. The close can't proceed on an unexplained gap.

  3. Total with shown work and compare to last month

    Roll up the verified data into category totals with the math shown, then run the month-over-month delta and the spend audit together — so the totals are checkable and the anomalies surface in one pass.

    you ask
    From the clean, reconciled data, total spend by category and show the formula for each. Then compare to last month's totals — biggest movers in dollars and percent — and scan for anomalies: charges far above normal, duplicates, and first-time vendors. Tell me the single line most worth explaining to leadership.

    what you get back A totals table with shown work ("Software 14,650 = sum of Category='Software', 41 rows"), a delta table ("Software +8,400 / +134%"), an anomaly flag ("Cloud Co 14,200 one-time overage"), and the one callout — Audit the month's spend folded into the close.

    Spot-check one category total by hand against the raw file. If it ties, you can trust the method across all of them — that single check is what makes the whole close defensible.

  4. Update the runway model

    Close isn't just backward-looking. Feed the verified actuals into the Build a cash-flow model you understand system so the close ends with a current, honest runway number, not a stale one.

    you ask
    Update cashflow-model.md with this month's actual closing balance and any changed recurring costs. Re-project 6 months, explain any formula that changed, and tell me the current runway and the first month the balance goes negative. Run one downside scenario: revenue flat, no growth.

    what you get back A refreshed model anchored to real actuals: "Closing cash 84,000; runway ~3.5 months (negative in month 4); flat-revenue downside: month 3." The forward number is now built on this month's verified close, not last month's guess.

  5. Write the summary and assemble the close pack

    End with the Brief leadership in plain English system — the narrative on top, the verified appendix below — and bundle every artifact so the whole close is one reviewable, repeatable pack.

    you ask
    Write the leadership summary under 200 words from these verified numbers — what changed, why (use this context: [one-off tax bill, planned migration]), and two things to watch — with an appendix tracing each figure to its file and formula. Then list the full close pack: clean files, reconciliations, totals, model, and summary. Save it all so next month is a re-run, not a rebuild.

    what you get back A under-200-word briefing whose every number traces to a verified step, plus a close-pack index (clean exports, reconciliations, totals-with-work, updated model, summary). The close is done — and every figure in it is defensible.

make it your own
  • Then present it to the board: once the close is verified, Assemble the board & investor finance pack turns it into the board story — the close makes the numbers right; the pack makes them presentable and defensible, over the exact same verified files.
  • Quarter-end or year-end: run the same arc over three months of data; lean harder on Audit the month's spend and add a quarter-over-quarter comparison to the totals step.
  • Tighten the recurring parts (Power Track): once stable, wrap the profiling and reconciliation steps in custom commands and let a scheduled agent (see the Capabilities tab) prep the clean files and exception lists overnight — so your morning is review and judgment, not data wrangling.
  • Lighter monthly cadence: if a full close is overkill mid-quarter, run just Make a messy export trustworthy → Audit the month's spend → Brief leadership in plain English for a fast pulse-check.
watch out for
  • Don't skip the cleaning step to save time. A close built on an unprofiled export is fast to produce and impossible to defend when a number is questioned — the early steps are what make the late ones trustworthy.
  • Every reconciliation must tie out and every total must show its work before it reaches the summary. A close where the numbers are forced to balance is worse than one that honestly flags a gap — never let Claude paper over an unexplained difference.
  • Claude runs the workflow; a human owns the close. You sign off on every number, you decide what's a write-off versus a chase, and confidential financial data stays in your approved workspace with anything truly private swapped for [placeholder] — the close is yours to defend, not Claude's.

you'll end up with A complete, repeatable month-end close in one workflow — clean inputs, reconciled accounts, totals with shown work, a month-over-month comparison, an updated runway, and a leadership summary — where every number is verified, traceable, and defensible.

Questions people ask

How long does a full month-end close take with this system?
Plan about a half-day for a full close — clean and profile each export, reconcile the accounts, total with shown work, compare to last month, update the runway model, and write the leadership summary. Once stable, the recurring data-wrangling can be prepped overnight so your morning is review and judgment.
Can I skip the cleaning step when a file looks fine?
Don't — that "clean-looking" file is usually the one that bites you. A close built on an unprofiled export is fast to produce and impossible to defend when a number is questioned; the early profiling and reconciliation steps are what make the late ones trustworthy.
What makes the close defensible to leadership or an auditor?
Every reconciliation ties out and every total shows its work before it reaches the summary, and each headline figure traces to a specific file and formula in the appendix. Spot-check one category total by hand against the raw file — if it ties, you can trust the method across all of them.
What if a reconciliation doesn't balance?
Flag the gap honestly — never let Claude paper over an unexplained difference. A close that forces the numbers to balance is worse than one that surfaces a real discrepancy, because the gap is exactly what you needed to find before signing off.