How to Write a Monthly Business Review Using AI

What a monthly business review must answer

A monthly business review answers four questions, in order: how did we do against target, why, what could go wrong next, and what do we need leadership to decide. That’s the whole monthly business review format. Everything else belongs in the appendix.

A monthly business review (MBR) is a recurring meeting, usually backed by a short written review or deck, that reports performance against targets, explains what drove the gaps, flags the risks ahead, and asks leaders for the decisions needed next month.

The Carry-Forward Review in brief

Most MBR templates stop at those four sections, which is why so many reviews feel like a rerun: the same risk shows up three months in a row, and nobody asks what happened to last month’s request. The Carry-Forward Review approach we teach at PromptPeakAI adds two bookends to the four sections.

• Open by settling last month. Every risk and ask from the previous review gets a status: decided, still open, landed, averted or dropped.
• Close by carrying forward. Every open risk and new ask becomes a line on next month’s opening list, with an owner and a date.

Operations leads who run good MBRs will tell you the settle step is where credibility comes from. Leaders forget what you asked for. They remember whether you followed up.

MBR, weekly and quarterly reviews

• Weekly: progress, blockers and next steps for your own team. That’s the job of a weekly status report, not an MBR.
• Monthly: performance against target, often cross-functional, built around decisions. That’s this guide.
• Quarterly: strategy, outlook and the bigger bets, often feeding a board report or a client-facing quarterly review.

Prepare your raw numbers and notes first

A monthly business review using AI can only be as good as the input you hand the tool, so spend ten minutes on an input pack before you open any tool. Four inputs do most of the work.

Build a clean metrics export

Keep it boring: one row per metric, with columns for Metric, Target, Actual and Prior month. If your KPIs change every month, fix that first; choosing the right KPIs once saves you rebuilding the review structure every cycle. If the export arrives full of merged cells and subtotals, clean the messy export before the AI sees it.

Upload the file itself, never a screenshot. OpenAI’s help page on data analysis warns that ChatGPT may not reliably extract exact values from image-based tables or scanned files. In practice, that’s where a lot of wrong numbers in AI-drafted reviews start.

Collect notes before you prompt

Your notes are where the drivers live. Keep a running doc through the month with one dated line per event: a big order slipped, a system went down for a day, two people left. It takes seconds each time, and it’s the difference between “cause unknown” and a driver leaders believe.

How to write a monthly business review using AI

Here’s the whole build in six steps. The order matters more than the tool: numbers first, words second, summary last.

1. Settle last month’s risks and asks against this month’s notes.
2. Have the AI compute every variance in code.
3. Draft performance, drivers and risks as separate sections.
4. Write the asks yourself, then let AI tighten them.
5. Write the executive summary last.
6. Reconcile every number against the source file before sending.

Upload last month’s review and paste this month’s notes. The AI’s job here is bookkeeping: match each old item to a status, and admit when it can’t.

You are helping me prepare this month's business review.

Attached is last month's review. List every risk and every ask it raised.

For each one, give:
- What it said, in one line
- Its status this month: Decided, Still open, Landed (the risk happened), Averted, or Dropped
- The evidence for that status, quoted from my notes below

If my notes don't show a status, write "No update — check with [owner]". Do not guess.

My notes on what happened this month:
[paste your notes]

Most guides get this step backwards. They tell you to upload the data and ask for “insights”, which hands the story to the tool before you’ve checked a single figure. Ask for calculation only, no commentary, so you can verify the table before any sentence depends on it.

Attached is our [month] metrics export for [team or business unit]. Columns: Metric, Target, Actual, Prior month.

Using code, not mental arithmetic, build one table with:
- Variance to target (absolute and %)
- Change vs prior month (absolute and %)
- A flag for any metric more than [5]% off target

Sort by the size of the gap in business terms, not by percentage. Include a totals row so I can check it against my file. Do not write any commentary yet.

Feed the checked table and your notes back in and ask for performance, drivers and risks. Leave the asks blank for now; they get their own treatment further down.

You are drafting a monthly business review for [audience, e.g. the operations director and the finance lead].

Use ONLY the variance table above and these notes:
[paste your notes]

Write three short sections:
1. Performance: 3 to 5 bullets, biggest gaps first, every number taken from the table
2. Drivers: for each flagged metric, one line on the cause, quoting the note it came from. If no note explains it, write "Cause unknown"
3. Risks: up to 3, each with likelihood (high, medium or low), impact, owner, and the date we'll know

Add a fourth heading, "Asks", and leave it empty. I will write that section myself.

Plain language. No words like "strong" or "challenging" without a number behind them.

Directors read the top four lines and skim the rest, so the summary has to be the most accurate part of the document. Writing it last, from finished sections, stops it inventing a story the sections don’t support. For more on the format, see how to write an executive summary leaders act on.

Using the finished sections above, write a four-line executive summary for [audience]:
- Line 1: the headline result against target, with the number
- Line 2: the single biggest driver
- Line 3: the risk that most needs attention
- Line 4: the decision we need, with its deadline

Add no new facts. If a line needs something the sections don't contain, write [MISSING] instead.

Turn variances into drivers, not excuses

A driver is the specific, checkable reason a number moved. “Market conditions” is not a driver. “Two renewals slipped to October because legal hadn’t signed off” is.

Rank gaps by business impact

Percentages mislead. A 40% miss on a small line item matters less than a 3% miss on your biggest revenue stream, yet AI (and most people) will lead with the 40%. That’s why Prompt 2 sorts by the size of the gap in business terms. Explain the top three or four gaps properly and send the rest to the appendix.

Make every driver traceable

Each driver line should point back to something real: a dated note, an email, an incident ticket. If nothing explains a gap, the honest line is “cause unknown, checking with [owner]”. Most finance directors will tell you they forgive that far faster than a confident guess that falls apart at the first question.

Finance teams formalise this with a driver column in their variance tables, and our guide to explaining variances in plain language covers that method in depth. When several drivers interact, building a data narrative helps you order them.

Write risks and asks leaders can act on

Risks tell leaders what might go wrong next month; asks tell them what to do about it. Both fail the same way, by being too vague to act on.

A risk line leaders trust

Give every risk four parts, kept to one line:

• What could happen, in plain words
• Likelihood: high, medium or low
• Impact in money, time or customers, wherever you can put a number on it
• Owner and the date you’ll know whether it’s landing

If your organisation already uses RAG status indicators, map likelihood and impact onto them rather than inventing a second scale. Cap the list at three. A review with nine risks effectively has none.

Turn requests into decisions

The asks are the reason the meeting exists, and they’re the section you should write yourself. Then let AI turn your rough version into something a director can say yes or no to, and build next month’s carry-forward list while it’s there.

Here are my rough asks for leadership this month:
[paste rough asks, e.g. "need more help on the backlog", "need a call on the vendor contract"]

Rewrite each one as a decision request with:
- The decision in one sentence, starting with a verb
- Up to two options, plus my recommendation
- The deadline, and what happens if no decision is made by then
- The owner

Then write a list headed "Carry forward to next month": every open risk and every ask, one line each, with owner and the date to check it.

Worked example: from spreadsheet to review deck

Here’s the same monthly business review using AI, run end to end in two tools. The input pack works in any assistant that can read a spreadsheet. What differs is what comes out the other end: ChatGPT is strong on the analysis and charts, while Claude can hand back the finished file for your monthly business review presentation.

The ChatGPT route

Attach the metrics export and run Prompts 2 and 3. ChatGPT runs the calculation, shows you the table, and can turn any flagged metric into a bar or line chart for the deck. File uploads and data analysis are limited on the Free plan and included on Go, Plus and Pro, and OpenAI’s pricing page lists interactive tables and charts on Plus and Pro.

If this is your first time, start with how to analyse a spreadsheet in ChatGPT, and see creating charts with AI for chart prompts that hold up in a leadership deck.

The Claude route

Claude’s edge is the output. Anthropic’s help centre confirms it can create Excel, PowerPoint, Word and PDF files on every plan, including Free, with a 30MB limit per file. On Free, Pro and Max, turn it on under Settings, then Capabilities, then “Code execution and file creation”. On Team and Enterprise it’s on by default unless an admin has switched it off.

Once the draft is checked, ask for the deck in one line: “Turn this review into a six-slide PowerPoint: title, summary, performance, drivers, risks, asks. Put the full variance table on an appendix slide.” You get a .pptx you can open, restyle to your company template and present.

If you live in Microsoft or Google

• Copilot in PowerPoint can build a deck from a Word document, so write the review in Word first, using heading styles, then generate the deck from it. It needs a work Copilot licence; our guide to using Copilot in PowerPoint covers the setup.
• Gemini in Google Sheets opens from the Ask Gemini button and works in a side panel next to your data, where it can build charts and pivot tables, provided your Google Workspace or Google AI plan includes it.

Check every number before it goes out

Five minutes of checking protects the credibility you’ll need for next month’s asks. The rule is simple: no number leaves your hands unless you can point to the cell it came from.

Every no has a fix that takes minutes, not hours. The routine below catches the rest.

The five-minute reconciliation check

1. Compare the totals row in the AI’s table with your source file.
2. Spot-check the three biggest numbers in the performance section against the file.
3. Run the audit prompt below and fix every mismatch it reports.
4. Read the summary once more and confirm it adds nothing the sections don’t say.

Check the draft review below against the attached source file.

List every number that appears in the draft. For each one, give:
- The number as written
- The row or cell in the file it came from
- Match or Mismatch

Flag any number you cannot trace to the file. Do not fix the draft; only report.

Draft:
[paste your draft]

When the draft still looks wrong

What AI can’t do in your MBR

When you write a monthly business review using AI, the tool is good at the arithmetic and the first draft. It’s poor at the parts that carry your reputation, and pretending otherwise is how a review goes sideways in the room.

Handling confidential company figures

Monthly numbers are usually confidential, so check which account you’re in before you upload anything. OpenAI states that ChatGPT Business data is excluded from model training by default, but a personal account on a work laptop is a different arrangement, and other tools have their own business terms. Our guide to whether ChatGPT is safe for work walks through the settings.

Should you tell leadership?

If anyone asks, yes, and say what the AI actually did: it calculated the variances and drafted the text, and you checked every number and wrote the asks. Most leaders care whether the numbers are right and the asks are clear, not who typed the first draft.

The first month takes longest, because you’re building the input pack and testing the prompts. From month two, a monthly business review using AI mostly runs on rails: last month’s carry-forward list opens the review, the same six prompts run on a fresh export, and your time goes where it belongs: on the causes and the asks.