The Eight Questions in a First Stakeholder Interview

First stakeholder interview questions, built backwards from the six things a case assembled from public sources could not settle, and what each answer changes.

Numbers in this note carry source tags ([POLICY], [THREADS n=10], [MINE]). What the tags mean.

The case under this piece is the returns case published on this site: returns and refunds at a mid-size online retailer, built from the published returns policy, its help centre articles, ten public complaint threads I picked myself [THREADS n=10] and a walkthrough of the published journey as far as it goes without placing a real order [WALKTHROUGH]. No client, no engagement, and not one interview. Nobody inside that company has been asked anything by me or on my behalf. That is a strange place to write about interviewing from, and it is also the only reason this particular list is worth your time: a case assembled entirely from outside produces an exact inventory of what the outside cannot tell you. Six of those gaps sit in the register as open questions, each with a date and a fallback [OPEN]; five of the six carry an owner, and the sixth is open precisely because no public source names one. The eight questions below are those gaps, turned around and pointed at a person.

A first stakeholder interview is the session where the things you could not establish on your own become rules with owners and dates against them. Which makes the agenda a derived document rather than a checklist: you do not walk in with eight good questions, you walk in with the eight holes in your own draft, ordered so that the answer most likely to resize the work arrives while there is still time to use it.

In short

• The agenda comes out of your draft, not out of a list somebody published. Every claim you tagged as your own inference and every owner slot you left open is a candidate question.
• Write down what a yes and a no each change in the document, before the session. A question whose two answers lead to the same next step is a question you can skip.
• Three of the eight can resize or cancel the work. Ask those first, in case the session is cut short.
• An answer with no name and no date against it cannot be rechecked by anybody, including you, and that only becomes a problem months later when somebody asks who agreed to it.

Where does a good interview question actually come from?

Not from a list of openers. "What are your main pain points" gets you a paragraph you could have written yourself before the meeting, and it puts the work of structuring the conversation on the person you came to learn from.

It comes from the draft you wrote before the meeting. Build the document first from whatever is public: the policy, the help pages, the complaints, the part of the journey you can walk yourself. Tag every claim with where it came from, then read your own tags. Every [MINE] is a number you chose and nobody confirmed. Every [OPEN] is a slot you could not fill from outside. That is your agenda, and it has a property no generic list has: you can say out loud what you will do with each answer.

The scorecard check behind this is 05, open questions listed with an owner and a date. It earns a place among twenty-five checks because a question with nobody's name on it quietly becomes an assumption, and assumptions surface during build at the worst price you will pay all year.

There is a second effect. Turning up with "your published policy says partial refunds are permitted only above 40 EUR of order value [POLICY], and I could not find who owns that figure" is a different opening from "tell me about returns". The first one can only be asked by somebody who did the reading, and it asks for a decision right rather than for a briefing.

The eight questions, and the gap each one came from

This is the plan I would take into a first session on the returns case, if access appeared tomorrow. Every row on the left exists because something on the right could not be settled from outside. The plan assumes one session and no guarantee of a second [MINE].

Six rows come straight out of the open questions register. Two do not: question 5 comes from the reach column of the stakeholder map, where the honest entry was that no public source names a route to the person who can reverse a refund, and question 6 from the exception path in the as-is, where the branch exists and everything around it is invisible from outside. Both gaps sat in the artifacts for weeks without being written down as questions, which is its own lesson about where to look.

1. Rule or target

A commitment can be breached, so it needs a mechanism, an owner and a consequence. A target cannot be breached, so anything you hang off it is decoration. "Aims to resolve" [HELP] leaves both readings open and rereading it will not settle them. Ask early: BR-07, the escalation in REQ-015 and the published wording all move with the answer.

2. Who owns the number

Owner and review trigger are separate things and you want both. A rule with a name against it can be renegotiated; one without a name can only be worked around, quietly. Scorecard check 16 asks for the name and the date, check 18 for the event that would make the rule wrong. If nobody owns it, that is not a failed question, it is a finding, and it goes in the register in those words.

3. Can you pull the queue

The one that can cancel the work. Four of the ten threads I read ended with an unowned case, and the same sentence in my case says the frequency across all cases is unknown from outside [THREADS n=10]. If the export shows parked cases are rare and short, the failure mode was real and not worth a release slot, and saying so is the job. Ask for the export rather than an impression: an impression cannot be rechecked afterwards, including by the person who gave it.

4. Which clock, and what happens at the weekend

A duration means nothing until the event that starts it is named, which is scorecard check 13. Two clocks in one process, one in clock hours and one in working days, decide a case under one and pay it under the other. Whoever runs the rota can answer immediately, and it is written down nowhere.

5. Who can undo it

The head of the department is the obvious first interview and rarely the most useful one. The person who receives the returned goods can hand an item back and unmake a refund the office has already promised [POLICY] [HELP], and nothing public says they are ever in the conversation where the refund window is set [OPEN]. Ask for the route as well as the name: a calendar invite, their manager, the weekly returns call, ten minutes on the bay. Then ask what that person loses if the change goes through, which is scorecard check 08.

6. What people do while a case waits

My as-is has a branch where work stops, and the interesting part is whatever has grown around it: a spreadsheet somebody maintains, a message to a colleague, a call made out of hours. From outside you can see that cases end there [THREADS n=10], never what happens around them, and a workaround in daily use is part of the current process whatever the documents say. If the answer is that the case simply waits, write that down too.

7. Which release, and when does the decision expire

Scope questions asked without a date come back as assumptions. The register carries this one as 30 April 2026 with a fallback for the day after. The sentence worth agreeing in the session is that fallback, not the intention to decide soon.

8. How long do you keep it

The 90 days in my draft is mine: I chose it, found no basis for it anywhere public, and tagged it so nobody inherits it by accident [MINE, OPEN, OQ-01]. It is the one question here whose right answer is somebody else's job, and where guessing produces a document that looks finished and is quietly wrong. Legal owns it, and the fallback is the shortest defensible period, escalated rather than absorbed.

What if you only get fifteen minutes?

Plan the short version before you need it. Take three: question 3, then question 1, then question 5.

Question 3 first because it can end the project, and an early no costs less than a late one. Question 1 second because more rows in the register depend on its answer than on any other: BR-07 status, the escalation in REQ-015 and the published wording all move with it. Question 5 third because it is the only one of the eight whose answer arrives from outside the room you are in, and because it costs one sentence to ask.

The other five survive being asked by email. Those three do not, because each of them needs a follow-up in the same breath.

What do you write down while they are talking?

Three habits, all of them cheap, all of them worth more later than they feel at the time.

Record the wording, not your summary of it, for anything that will become a rule. "Aims to resolve within five working days" and "must be resolved within five working days" are one word apart and produce different documents. Your paraphrase always drifts towards the version that suits your draft.

Tag the answer with the session and the date. The pack's provenance convention is built for exactly this: add a tag of your own, [INTERVIEW 2026-05-12], in the same line as the claim. The worked case carries no such tag anywhere, because it had no interviews, and that absence is visible on every page. An unattributed answer is a fact nobody can check, which is what the convention exists to prevent.

Write down what you did not get. The vague answer is more useful in your notes than the clear one, because it is the one still open in six weeks. Same three columns as the register: owner, date, and what happens if the date passes.

The review log on this case records two sessions with peer reviewers, on 2 and 6 April 2026, five objections in total. Neither reviewer works for the retailer and neither session was an interview with anybody inside it, which the log says on its face. Keep the two kinds of session apart in your own notes, because a reader who cannot tell them apart has to assume the weaker of the two.

The blank version

Copy the table, fill in the middle column first, and only then write the questions. Working in that order stops you from importing somebody else's agenda into your own document.

Two rules for filling it in. If columns four and five say the same thing, cut the row: the answer changes nothing, so the question is curiosity rather than analysis. If column six is empty, you do not yet know what the answer is for, and asking it will produce a note rather than a decision.

Then order the rows by which answer would change the most, and mark the top three. Those are the ones you ask when the meeting turns out to be fifteen minutes.

Where to start

Take the document you have now and read only your own tags. The inferences you made and the slots you left open are your first draft agenda, and it is more specific than any published list of interview questions, including this one, because it is about your case.

Then run the artifacts through the scorecard: twenty-five yes-or-no checks, free, no email address asked for, and check 05 is the one this habit answers to. The worked case and the six blank templates are in the Proof Pack. If your stakeholder map is still a list of job titles, the map that survives contact with the project is the piece to read before the session, and the questions a reviewer asks is what happens to your notes afterwards.

Six of the eight above are the open questions in my register, all eight are unanswered, and they will stay that way: the case is built from public sources on purpose and nobody at that retailer is going to be asked. That is the price of a case anybody can check. With access you can close in one session what I cannot close at all, and having closed it, with a name and a date beside the answer, is most of what a reader is looking for.

All field notes · The portfolio guide · More in The Case File