TL;DR
Paste this week’s numbers (this week / last week / target), notable events, and your audience into ChatGPT, Claude, or Gemini, and ask for a fixed structure: a 2-3 sentence BLUF, three wins, three risks, and one specific ask. That compresses a 60-90 minute write to about 15 minutes. The model writes the narrative; you supply every number and spot-check three of them, because AI rounds and miscopies. Keep the whole thing under 250 words on one screen. The prompt below is copy-ready.
The task
Every Monday you write the weekly KPI summary for your team or leadership. Pulling the numbers is quick; the 60-90 minutes goes into wordsmithing the narrative. A good summary needs a TL;DR a busy exec will actually read, two to four wins worth celebrating, two to four risks that need attention, and one ask — the specific thing you need from leadership to move faster. AI can take that writing from 90 minutes to roughly 15 if you feed it well. The catch: the model is only as accurate as the numbers you paste, so the human job shifts from writing to data hygiene and judgment.
When AI is the right tool
- You write a summary in the same format every week.
- The metrics are stable week to week and the audience is the same.
- You have the numbers cleaned and ready to paste.
- You want a consistent voice across weeks, or across teammates writing on rotation.
When not to lean on AI alone
- Board-ready summaries that get scrutinized line by line. Use AI for the first draft, then finalize yourself.
- Weeks with a major outlier event (incident, launch, layoff). The framing is delicate and personal.
- Numbers that are sensitive or pre-public (earnings, M&A signals). Be deliberate about what you paste — see the data-privacy note below.
What to feed the AI
The output quality is set almost entirely by the input. Give it all five of these:
- This week’s metrics with last week’s value and the target as comparison columns.
- Notable events: launches, incidents, campaigns, holidays, churn spikes.
- The audience: who reads it, what they care about, and what they will ask in the meeting.
- House style: voice, format, and the kind of line that gets cut for being vague.
- Last week’s summary as a structural reference so the format stays consistent.
Copy-ready prompt
Write this week's KPI summary.
Audience: [who reads it, what they care about]
Voice: [3-5 adjectives, e.g. "concise, candid, action-oriented"]
This week's metrics (this week / last week / target):
- [metric 1: this / last / target]
- [metric 2: ...]
- [metric 3: ...]
Notable events this week:
- [launch, incident, holiday, campaign, etc.]
Output structure:
- BLUF / TL;DR (2-3 sentences max — a busy exec must get the week from this alone).
- Top 3 wins (each tied to the specific number that earned it).
- Top 3 risks (use the leading indicator, not the lagging one).
- 1 ask (a specific request: money, headcount, a decision, an unblock).
- 1 "what to watch next week" sentence.
Constraints:
- Total length under 250 words.
- Every claim must reference a number.
- No fluff verbs ("crushing it", "doing well"). If a metric is flat, say flat.
- Do not invent or recompute any number. Use only the figures I pasted.That last constraint matters: language models are unreliable at arithmetic, so you paste the math and the model only narrates it. If you need a computed figure (a week-over-week delta, a percentage), compute it in your spreadsheet first, or use a model with a code/analysis mode that runs real calculations rather than guessing.
Which model, and the privacy line
Any current frontier model handles this well. The differences that matter for a weekly summary are context window (how much history you can paste) and data governance (whether your numbers train the model).
| Model (as of June 2026) | Plan to use | Context window | Trains on your inputs? |
|---|---|---|---|
| ChatGPT (GPT-5.5) | Plus $20/mo | ~320 pages in-app on Plus; full 1M tokens only on $200 Pro | No on Enterprise/Team/API; consumer Free in the US carries ads since Feb 2026 |
| Claude (Sonnet 4.6 / Opus 4.7) | Pro $20/mo | 1M tokens standard | No on paid plans by default |
| Gemini (3.1 Pro) | Google AI Pro $19.99/mo | 1M tokens | No on Workspace/enterprise |
On the privacy line: ChatGPT Enterprise, Claude (Team/Enterprise), and Google Workspace all promise that your inputs are not used for training by default, and admins keep data ownership. If you are on a consumer tier or unsure of your org’s settings, the safe pattern is to paste deltas and percentages, not absolute figures — “MRR +4.2% vs target +6%” tells the story without exposing the dollar amount. For pre-public numbers (earnings, M&A), do not paste them into any external tool at all. Anthropic and OpenAI both document their data-handling defaults publicly; Anthropic’s privacy center is a good place to confirm what your specific plan does with inputs.
What a great KPI summary looks like
The format follows BLUF — Bottom Line Up Front, the military and management standard of leading with the conclusion, not the table of contents. A few rules that hold up across teams:
- One screen, under 250 words. If it scrolls, it stops being a summary.
- Lead with the bottom line. “Pipeline coverage dropped to 2.1x against a 3x target and that is next quarter’s risk” is a BLUF; “this report covers pipeline and activation” is a table of contents.
- Favor leading indicators in the risk section. Strategy teams that track roughly 60% leading to 40% lagging indicators catch problems before they show up in revenue. Trial-to-paid conversion is a leading indicator; churned MRR is the lagging confirmation.
- Keep the metric count tight. Analysis of more than 20,000 strategic plans found that lean frameworks of around 9-11 measures outperform sprawling 60-plus dashboards; a weekly summary should surface 5-7 numbers, not 30.
- Status at a glance. A red/amber/green marker per metric lets an exec scan the week in seconds.
How to check the output
- Read only the TL;DR. Does it tell the week’s story on its own?
- Spot-check three numbers against your source data. AI rounds and miscopies.
- Is the ask actually specific? “Support from leadership” is not an ask; “approve hiring 2 SDRs by Friday” is.
- Does it match your house voice, or did it suddenly start “leveraging synergies”?
Common mistakes
- Numbers without narrative. The narrative is the value-add; a wall of figures is just the dashboard reprinted.
- A vague ask. If you cannot answer “what changes if leadership says yes,” it is not an ask.
- Burying the lede. The biggest news belongs in sentence one, not paragraph three.
- The same wins every week. If “growth team shipped X” appears four weeks running, the audience tunes out. Rotate what you report.
After a month, audit the summaries
Once you have four weeks, paste them in together and ask the model: “What patterns repeat? Which metrics do I report that no one acts on?” You will catch yourself reporting numbers that never drive a decision — rotate or cut them, and add the ones the audience keeps asking about in meetings. This is also where AI earns its keep beyond the weekly draft: comparing your own output over time is tedious for a human and fast for a model.
FAQ
Should I let AI generate the metrics themselves? No. Paste exact numbers and tell the model not to recompute them. Language models are unreliable at arithmetic and will confidently produce wrong deltas. Do the math in your spreadsheet or in a model’s analysis/code mode that runs real calculations.
How do I handle confidential data? On a paid Enterprise, Team, or Workspace tier, your inputs are not used for training by default. If you are unsure, paste only deltas and percentages instead of absolute figures, and never paste pre-public numbers (earnings, M&A) into an external tool.
Can the AI suggest the ask? Yes, but you finalize it. The ask reflects political judgment about what leadership will actually grant and when — that is yours to own.
How long should the summary be? Under 250 words, on one screen. If an exec has to scroll, the format has failed. Five to seven headline metrics, three wins, three risks, one ask.
Which model is best for this? Any of GPT-5.5, Claude Sonnet 4.6, or Gemini 3.1 Pro handles a weekly summary well. Choose on context window and data governance, not raw capability — see the table above.
Related
- AI writes KPI narrative
- AI writes weekly KPI report
- AI writes dashboard takeaways
- Write the A/B Test Summary With AI
- KPI Commentary Prompts: 12 Templates for Numbers That Tell a Story
Tags: #Data analysis #Workflow #Research #KPI