What this covers
The pain: a 12-tab spreadsheet that took two days to build, presented to a VP who looks at it for 30 seconds and asks “so what?”. This workflow turns the table into a four-sentence narrative that survives the 30-second test, with the table attached as evidence. The trick is doing the pivoting and aggregation yourself first, then handing AI a compact summary to narrate — not the raw 5,000 rows.
Who this is for
Analysts, PMs, finance, and ops — anyone who explains numbers to people who do not live in spreadsheets. Especially useful for weekly KPI emails, board pre-reads, and any meeting where the table is on the screen but the audience needs sentences. Not for analysts presenting to other analysts, who actually want the raw table.
When to reach for it
Weekly KPI reports, quarterly business reviews, ad-hoc deep dives where leadership wants the “headline”, investor updates, and post-launch debriefs. Also useful when you inherit a recurring report from someone else and need to learn what the numbers usually do — AI narrating six weeks of historicals is faster than reading them yourself.
Before you start
- Decide who the reader is: exec (wants the verdict), peer team (wants the story), or your own future self (wants the anomaly).
- Aggregate raw data first — group by the right dimension (week, segment, region) and limit to under 50 rows. Anything bigger and AI hallucinates totals.
- Strip PII, customer names, internal codenames before pasting. Replace with role tags like
Customer AorEU segment. - Have one reference number in mind that you already know is true. AI will sometimes invent a percentage — having a ground-truth lets you catch it.
- Know the comparison frame: vs last week, vs plan, vs same period last year. AI will pick a flattering one if you do not specify.
Step by step
- Pivot or aggregate the raw data first. Do not feed 5,000 rows to AI — pivot it to under 50 rows of weekly or segment-level totals first.
- Export the pivot as a compact markdown table or paste it as CSV with headers. Include row totals and column totals if they exist.
- Prompt AI with explicit framing. Example:
Audience: VP of Sales, 30 seconds, decision-oriented.
Comparison: vs prior 4-week average.
Output: 4 sentences. Cover (1) headline, (2) biggest positive, (3) biggest miss, (4) one anomaly to investigate.
Do not editorialize. Do not use "leverage", "synergy", or "momentum".
Table:
[paste table here]- Review the narrative line by line. Cross-check every number, percentage, and direction against the table. AI confuses “up 12%” with “up 12 percentage points” routinely.
- Edit for voice. AI defaults to a generic professional register. Replace one or two sentences with how you actually talk in Slack.
- Pair the narrative with the table in the final deliverable. The narrative leads, the table is evidence. Do not delete the table — execs check it after they read the narrative.
First-run exercise
Pick a report you already wrote by hand last week. Run the same pivot through AI and compare. Mark each AI sentence as: factually correct and useful, factually correct but generic, or factually wrong. Most analysts find AI gets the headline right but invents anomalies — there is no “spike” if the segment only has three data points. For the second run, change only the audience tag in the prompt; keep the table and the comparison frame the same. You should see the verbs and the lead sentence shift, not the underlying analysis.
Quality check
- Number audit: pick three numbers from the AI narrative at random and trace them back to a table cell. If any does not match, the whole draft is suspect.
- Anomaly test: does the “anomaly” AI surfaced actually deviate by more than two standard deviations or a meaningful business threshold? If not, drop it.
- Comparison frame stated: every percentage must say “vs what”. A 12% increase vs last week, vs last year, and vs plan are three different stories.
- Decision test: after reading the narrative, can the reader name one action they would take? If not, you described instead of explained.
- Privacy sweep: no customer name, no internal codename, no individual employee should appear by name unless it is a public press item.
How to reuse this workflow
- Save three prompt templates: weekly KPI narrative, quarterly review narrative, anomaly investigation. Each is the same skeleton with a different audience tag.
- Standardize the pivot shape across reports — same dimensions, same comparison columns. AI gets dramatically more consistent when the table shape stops changing.
- Keep a
numbers-AI-got-wrong/log. Patterns emerge: AI confuses percentage point vs percent, rounds away small denominators, infers direction from absolute values. Add these to the prompt as explicit rules. - Every quarter, rerun an old narrative against the new model. If the headline still holds, your prompt is robust; if not, examine which sentence type drifts.
- Share the prompt with one colleague who reads many reports. They will tell you which sentence to drop.
Recommended workflow
Raw data → pivot to under 50 rows → compact markdown table → AI narrative with explicit audience and comparison frame → number-by-number fact-check → voice edit → pair narrative with table as evidence.
Common mistakes
- Feeding raw transactional data — AI invents aggregates because it cannot do real math on 5K rows.
- Not fact-checking AI numbers — at least one number per draft is silently wrong.
- Letting AI invent anomalies that are not statistically real, especially on small segments.
- Skipping the comparison frame — “revenue up 12%” with no baseline is meaningless and reads as spin.
- Replacing the table with the narrative — executives lose trust when they cannot verify.
- Reusing the same narrative voice for exec, peer, and ops audiences. The headline changes per audience, not just the wording.
FAQ
- My data is sensitive. Can I still use AI?: Yes, with two precautions: aggregate before paste (no row-level data), and replace identifiers with tags. The narrative does not need to know it was “Customer Acme” — “Customer A” works.
- Why is AI getting percentages wrong?: Most commonly because the table has implicit percentages mixed with absolute values. State explicitly in the prompt which columns are absolutes and which are already percentages.
- Can AI find anomalies for me?: It can flag candidates but it will over-flag on small samples. Treat AI anomalies as “look at this” not “this is real”.
- Should I just use a chart instead of a narrative?: Use both. A chart shows shape, a narrative gives the verdict. Executives want the verdict first.
- What model size is enough?: A mid-tier model is fine for under 50-row tables and 4-sentence narratives. Larger models help when the table has more than 10 columns.
- How do I prevent the narrative sounding like AI?: Add a sentence in your own voice as the lead, and write the action recommendation yourself. Those are the two highest-signal sentences.
Related
Tags: #Tutorial #Productivity #Excel