Academy · AI in analysis · Guide
AI may explain your business numbers,
but it must never make one up.
AI assistants summarise tables, suggest follow-up questions and draft reports. With business numbers the line is sharp: every statement has to trace back to a number someone can recompute. This guide shows what AI is good for in analysis, what it isn't, and how to check.
Which job would you give an AI with your numbers?
2 questions in the plan. First question: “Summarise this quarter's income for the partners' meeting.”
May and may not
Calculate first, explain second Trace every statement to a cell
“Summarise this quarter's income for the partners' meeting.”
Explain, yes. Invent, no.
AI assistants summarise tables, suggest follow-up questions and draft reports. With business numbers the line is sharp: every statement has to trace back to a number someone can recompute. This guide shows what AI is good for in analysis, what it isn't, and how to check. AI may summarise a table everyone can see, suggest follow-up questions and draft a report. It may not quote a number that appears in no evaluation, present a cause as fact or make a decision without a person.
From reading to doing
Your questions. For your numbers.
Prepare an evaluation.
2 questions selected
- “Summarise this quarter's income for the partners' meeting.”
The AI works only with the table everyone can see.
- “Why have Amazon refunds gone up?”
A guess stays a guess until an evaluation supports it.
Calculate, explain, check.
but it must never make one up.
Three steps in which the calculation always comes before the explanation and a person has the final word.
“Summarise this quarter's income for the partners' meeting.”
The AI works only with the table everyone can see.
Calculate first, explain secondThe number comes from a traceable evaluation, such as a pivot with source, period and aggregation. Only then may an AI put it into words. A number that exists only in the AI's text does not exist.
Trace every statement to a cellRead the AI text sentence by sentence and ask: which cell does this come from? Phrases like "probably because of the campaign" are guesses and have to be marked as such.
People decide, with a trailAn AI may suggest what to look into. A person decides what happens, and the decision links back to the evaluation, so you can later see what it rested on.
Maysummarise a visible table · "Largest channel in October, per the pivot …"
Maysuggest follow-up questions · "What does it look like without refunds?"
May notquote numbers without a source · "Revenue is up by about 20%"
May notpresent a cause as fact · "… because of the new campaign"
The playbook
Calculate, explain, check.
AI in an evaluation
but it must never make one up.
Calculate first, explain second → Trace every statement to a cell → People decide, with a trail
- 01
Calculate first, explain second
The number comes from a traceable evaluation, such as a pivot with source, period and aggregation. Only then may an AI put it into words. A number that exists only in the AI's text does not exist.
Example · Willowcourt Advisory (fictional)“Summarise this quarter's income for the partners' meeting.”
AI draft from exactly that table - 02
Trace every statement to a cell
Read the AI text sentence by sentence and ask: which cell does this come from? Phrases like "probably because of the campaign" are guesses and have to be marked as such.
Example · Nordwind Travel Goods (fictional)
AI suggests possible reasons as questions“Why have Amazon refunds gone up?” - 03
People decide, with a trail
An AI may suggest what to look into. A person decides what happens, and the decision links back to the evaluation, so you can later see what it rested on.
Example · Harbourview Agency (fictional)“Should we hand a client board to a second team?”
AI lists the arguments for and against
Your quality check
Will the evaluation hold up?
Tick what you have checked. Your selection stays in this view only.
0 of 3 points checked
What this article is based on.
Editorial page with fictional examples (Willowcourt Advisory (fictional), Nordwind Travel Goods (fictional), Harbourview Agency (fictional)). The guide describes an approach; where Peak Analytics covers a step today, the page says so explicitly. No measurements, no customers, nothing is sent or saved.
- Editorial demo
Synthetic numbers in the browser. No real data is read, computed or saved.
- Checked evidence
Peak Analytics is a preview: pivot over orders, receipts, tasks, goals and files; count, sum, average, minimum, maximum; day to year; CSV export; sharing a released snapshot; largest value as a task.
peak-one app/w/[scope]/apps/analytics, lib/bi/model.ts, lib/views/pivot.ts, lib/bi/share.ts · checked 2026-10-29
The calculation first, the words after.
If nobody can recompute a number, don't trust it just because it is well phrased.
Frequently asked questions
Do I need Peak Analytics for this guide?
No. The guide works with any spreadsheet. Peak Analytics is a preview; where it covers a step today, the page says so.
May an AI write reports about our numbers?
Yes, as a draft from a visible evaluation. Every number in the text must trace back to a cell, and a person checks before it goes out.
How do I spot an invented number?
It appears in no evaluation. Ask for the source of every number; if there is none, the number doesn't count.
Does Peak Analytics have AI?
Not yet. AI questions on data, AI explanations and AI briefs are planned. Today the preview calculates without AI, every number traceable in the pivot.