← All Kits · All Guides

The Analyst Request Intake: Six Questions Before You Open the Data

Michael Nocito · Updated August 2026 · Every number on this page was worked before it was published

"Can you pull the numbers for the North region?" Two days later the answer is wrong, not because the query was wrong, but because North meant the sales territory and you used the postcode grouping, and it was needed for a board paper that went out yesterday.

What you do: six questions, asked in five minutes, before you open anything. Then one paragraph back in writing.

The short version. The cheapest analysis to fix is the one you have not started.

The six questions

1. What decision does this feed?

Not what do you want to see, but what happens next. "We are deciding whether to keep the Tuesday delivery slot" tells you the population, the period and the precision needed. "Just interested" is a legitimate answer, and it correctly sizes the work at an hour rather than a week.

The test question when the answer is vague: what would you do differently if the number came back twice as high? If nothing, this is a report and not an analysis. That distinction is worked in report against analysis.

2. Who reads it, and what do they already believe?

An operations lead who knows the data wants the detail and the caveats. A board wants one sentence and a chart. And knowing what the audience currently believes tells you whether your result is a confirmation, which needs little defence, or a contradiction, which needs a great deal.

3. What exactly does the measure mean?

This is where the days are lost. Every term needs a definition somebody would sign:

Ask for the definition they use, not the one that seems right. If two people give different answers, you have found the real deliverable, and it is worth more than the number. Defining metrics goes further into this.

4. What period and what population?

Include or exclude: cancelled orders, internal accounts, test records, refunds, partial months, the acquisition that closed in March. Each of these silently changes a total by a few percent, which is exactly the size of gap that produces a long meeting.

5. What number do you expect?

The most useful question on the page and the one that feels strangest to ask. If they say around 400 and you get 1,800, one of you is using a different definition and you now know before publishing rather than after. If they have no expectation at all, that is information too: nobody has looked at this before, so budget time for the data being messier than anyone assumes.

6. When is it needed, and what is the smallest useful version?

Deadline first, then scope to it. Ask what could be dropped if time ran short. People are far more willing to name the optional parts in advance than to accept them missing on the day.

The confirmation paragraph

Send this before starting. One paragraph, no template, no form:

To confirm what I am building:

Question:    Did the Tuesday delivery slot lose volume after the June change?
Measure:     Delivered orders, counted at delivery date, excluding cancellations
Population:  North sales territory, external accounts only
Period:      Jan to Aug 2026, comparing Jan-May against Jun-Aug
Output:      One chart plus three lines of commentary, in the ops pack
Needed by:   Thursday 4pm
Assumption:  Internal test accounts (prefix ZZ) are excluded

Shout if any of that is wrong, otherwise I will start this afternoon.

It takes four minutes. It converts every later disagreement from "you got it wrong" into "we should change this", which is a different conversation and a much better one.

The three answers that change the job

How to apply this to your own work

1. Put the six questions somewhere you can see them during a conversation, not in a document you will open afterwards.
2. Send the confirmation paragraph on the next three requests, whatever their size, and see how many get corrected.
3. Keep a note of every definition you agree. After a few months that note is a metric dictionary nobody had to be funded to write.
4. Record the expected number and compare it to your result. Where they differ, lead your write-up with the reason.
5. When a request changes, restate the paragraph rather than absorbing the change quietly.

The one habit to keep

Ask what decision the number feeds. It is the question that most often reveals the request should be a different request, and it is the one that stops good work being spent on the wrong question.

On your current piece of work, could you write that paragraph today without asking anybody?