Month-end variance commentary and board-pack drafts
Turn your closed actuals into a first-draft variance narrative and board-pack pages your team owns and edits, instead of writing every line from a blank page each month.
Tools you'll use
Month-end variance commentary is the written narrative a finance team produces after close that explains how actuals moved against budget or forecast, why, and what it means, then packages it into the board pack. Today someone opens a blank document and writes the same thing every month: revenue was up or down, here is why, here is what we are doing. The numbers come from the close. The hard part is the writing, chasing department heads for explanations, and formatting it into pages a board can read in ten minutes. That work lands in the tired days right after close, and the cost is real and recurring: at least three-quarters of finance teams spend five to six hours a week recreating financial reports, up to 300 hours a year (insightsoftware 2024 Finance Team Trends Report). The median organisation also takes 6.4 calendar days to close its books (APQC), so the commentary deadline is already tight when this work begins.
This use case points an AI tool at your reconciled actuals, your budget or forecast, and last month's commentary, and asks it to produce a first draft. Not the final word. A draft that already decomposes the big variances, flags what crosses your materiality threshold, drafts the narrative in your house style, and lays out the standard board pages. Your analyst then edits, corrects, and adds the judgment a machine cannot have.
It matters because the time saved is real and recurring. The blank-page problem disappears. The team reviews and sharpens instead of typing. And because you build the automation yourself with one of these tools, you own it. There is no vendor in the loop, no per-seat SaaS for this, and you can change the template or the rules whenever the board changes what it wants to see.
Moriva's take
Gate 1, real work: this attaches to the monthly close, a workflow your team runs every single month with a hard deadline, so it clears easily. Gate 2, owned: an analyst can stand up the draft generator in a focused week with Claude Code or Codex, then run and tweak it without us. Gate 3, measured: count the hours from "close done" to "commentary drafted" before and after, and the saving is obvious within two cycles. The reason this is CAREFUL not GO is the data is financial and pre-publication, and a confident-but-wrong number in a board pack is a real problem, so a human reviews and signs off every figure before it leaves Finance.
How do you month-end variance commentary and board-pack drafts?
- 1
Pull your last three board packs and write down the rules in your head
Before any tool touches data, gather three recent months of commentary and the board pack template. Note the unwritten rules: your materiality threshold (for example, flag anything over 10 percent or 50k, whichever is larger), the order pages appear in, the tone, and which lines always get explained. This becomes the brief you hand the tool. A tool can only match your standard if you have made the standard explicit.
- 2
Stage the inputs as plain files the tool can read
Export the closed actuals, the budget or latest forecast, the prior-period actuals, and last month's commentary into a folder as CSV or Excel files. Keep the chart-of-accounts mapping in there too. Use sanitised or internal-only data, not anything customer-identifiable. The tool reads what is in the folder, so a clean, consistent folder each month is half the job.
- 3
Have Claude Code or Codex build the variance engine
Open Claude Code or Codex in that folder and describe the task in plain English: compute actual minus budget and actual minus prior period for every line, flag everything past the materiality threshold, and decompose the big movements into price, volume, and mix where the data allows. It writes and runs a script that produces a clean variance table. You own that script, so you can read it, correct a formula, and rerun it yourself.
- 4
Add the commentary draft against your house style
Point the tool at last month's commentary and your three sample packs, and ask it to draft this month's narrative in the same structure: what happened in business terms, why, the impact on the rest of the year, and the recommended action. Tell it to write only from the numbers in the variance table and to leave a clearly marked gap wherever it needs an explanation a human must supply. This stops it from inventing reasons.
- 5
Generate the board-pack pages from a fixed template
Have the tool fill your standard board pages: the summary first two pages, the budget-versus-actual tables with both dollar and percentage columns, and the supporting charts. Keep the template fixed so output is consistent month to month. For non-coders doing the assembly and wording, Claude Cowork can take the variance table and draft the pages directly without writing scripts.
- 6
Review, correct, and route the gaps to department heads
An analyst reads every flagged variance, checks each number against source, and fills or corrects the explanations. Where the draft marked a gap, send a short, specific question to the relevant department head rather than a vague request. This is where human judgment lives and where you catch anything the tool got wrong. Nothing leaves Finance unreviewed.
- 7
Lock the monthly run and time it
Once two cycles look right, write the steps down as a short runbook so anyone on the team can run it. Record the clock time from close-complete to draft-ready, and again to board-pack-ready. Compare against the old baseline. That number is your proof and your case for keeping or extending the automation.
What could go wrong (and how to handle it)
The tool invents a plausible-sounding reason for a variance that is actually wrong.
Instruct it to write only from the numbers and to mark a visible gap where it lacks a real cause. The analyst fills those gaps from department heads. Treat every causal claim as unverified until a human confirms it.
A wrong number reaches the board because nobody traced it back to source.
Require a line-by-line tie-out of every figure in the pack to the close before sign-off. Keep the variance script readable so the math can be audited. The draft is a starting point, never the final record.
Sensitive, pre-publication financials are exposed.
Use internal-only data and follow your firm's policy on where company data may be processed. Keep customer-identifiable detail out of the inputs. Confirm the tool's data handling meets your controls before the first real run.
Over-automation: the team stops thinking and just forwards the draft.
Keep the analyst review as a hard, non-skippable step. The tool removes the typing, not the judgment. Make clear that the person who signs the pack owns every word in it.
Running variance analysis on late or unreconciled actuals produces confident nonsense.
Only run the generator after the close is complete and reconciled. Garbage in is garbage out no matter how good the draft reads. Gate the automation behind the close sign-off.
Commentary drifts from board expectations or compliance language over time.
Re-feed the tool the latest approved pack each quarter so it tracks current style. Keep a human owner of the template. Review the tone and disclosures whenever the board or auditors change what they expect.
Prompts to get started
FAQ
Will this let the AI sign off the board pack on its own?
No. It produces a draft. A human reviews every number, ties it to source, fills the gaps the tool marked, and signs off. The point is to remove the blank-page and typing time, not the accountability. The person who presents the pack still owns every line.
Is it safe to put our financials through one of these tools?
That depends on your firm's data policy, which you should confirm before the first real run. Use internal-only data, keep customer-identifiable detail out, and check the tool's data handling against your controls. Because you build and run the automation yourself, you decide where the data goes rather than handing it to a third-party SaaS.
We do not have engineers. Can we still do this?
Yes. Claude Cowork is built for operators who do not code and can take a variance table and draft the commentary and pages directly. If you want the automated number-crunching script as well, Claude Code or Codex can build it in plain English, and one analyst can usually stand it up in a focused week.
How is this different from buying an FP&A platform with AI built in?
A platform is a subscription you rent and cannot change. This is an automation you own. You can read the script, fix a formula, change the template the day the board asks for something new, and run it without anyone's permission. No per-seat cost for this workflow and no vendor in the loop.
What is the realistic time saving?
Most teams measure the clock from close-complete to draft-ready. The blank-page and first-draft work, which is often a day or more, drops to the time it takes to review and correct. Measure your own baseline over two cycles and compare; the saving is visible quickly and recurs every month.
Sources
- At least three-quarters (75%) of finance teams dedicate a minimum of five to six hours each week to recreating financial reports, equating to up to 24 hours a month or 300 hours per year. — insightsoftware 2024 Finance Team Trends Report (conducted with Hanover Research)
- At the median, organizations need 6.4 calendar days to close out a month's books (APQC Open Standards Benchmarking, ~2,300 organizations). — APQC, via CFO.com, 2023
More from Finance & Accounting
Reconcile accounts faster at close
Use an agentic tool to match transactions, surface only the real exceptions, and draft reconciliation write-ups every close cycle, so your team shrinks a multi-day chore into a few reviewed hours.
Expense categorization and anomaly flags
Build an owned script that reads your card and expense exports, assigns a GL category to each line with a confidence score, and flags the odd ones (duplicates, out-of-policy amounts, unusual vendors) for a human to check.
Draft and review journal entries
An AI agent drafts recurring and adjusting journal entries from your source data and prior periods, attaches the support, and flags anomalies for a human to approve, cutting the manual grind out of month-end close.
Want help shipping this?
We'll build it with your team on your real work — and leave you owning it, not renting it.