Stakeholder Interview Guide Template

First, list the three to eight things the interviews have to settle. Take them from the signed scope or the sponsor's words at the kickoff meeting.

Turn that list into 8 to 12 numbered questions, each answerable by more than one person. For each, list the roles that hold a view worth having, count the people in them and decide how many you will ask. That number is the question's target.

If only one person can answer a question, cover it in a separate briefing. If nobody can, leave it to a document review. Not sure who holds which views? Sketch a stakeholder analysis first.

Set every target before the first interview, because a target picked after the answers arrive tends to fit them. Federal survey teams plan ahead too. The US Office of Management and Budget calls it best practice to estimate a survey's response rate in advance and plan a nonresponse bias analysis if it is below 80%¹.

Nielsen Norman Group (NN/g), a user experience research firm, says stakeholder interviews are usually semistructured. Instead of a script, you use a discussion guide that gives direction but doesn't need to be followed precisely². Write one guide per role from the same numbered set, keeping every question's number. Mark in advance any question that doesn't apply to a role, which lowers its target. Every guide has six parts.

1. 1

   The consent note

   Read it, ask to record and wait for a clear yes.

2. 2

   Their role, in their words

   What they actually do, which tests the roles you planned.

3. 3

   One walkthrough

   The last time the process ran, start to finish.

4. 4

   The numbered questions

   This role's questions, each with a probe for an example.

5. 5

   The open question

   What they'd change, or what you should have asked.

6. 6

   Who else

   Who else to ask, and who sees it differently.

The walkthrough comes before the numbered questions for a reason. NN/g researcher Maria Rosala notes that if you ask reflective questions too early, you may get stereotypical answers³, because people haven't yet recalled real events.

Word each question so it asks one thing and doesn't hint at an answer. Version every guide from day one. If you add a question midway, give earlier interviewees a 5-minute follow-up rather than counting them as covered.

Start with two or three people who do the work; they show whether your roles are real and your questions land. Book directors and the sponsor once you can ask about something specific. If only a few people can answer a question, book them early, since nobody else can stand in.

Invite people with the sponsor's backing and promise no attribution by name. Share the purpose, not the question list, so you hear what happened rather than a prepared answer. Keep some slots free at the end: NN/g likes to close stakeholder interviews by asking "Who else should I speak to?"², which can lead to another key stakeholder.

Open every interview with a short consent note. Lead with who you are, who is paying, what it's for, how long it takes and that they can stop. The Common Rule, the US rule for federally funded research with people, is a good model even where it doesn't apply. It says consent must begin with a concise and focused presentation of the key information⁴.

Be specific about who sees what: themes with counts and quotes attributed by role, never a record of who said what. In a small team a job title can give someone away, so tell them at the time and ask how to use that answer. Set a deadline for taking an answer back, and ask nobody to waive their rights; the Common Rule bars that too.

After each interview, log one outcome per question in a coverage matrix: answered, could not say, not applicable, not asked or withdrawn. Mark answers as agreeing or dissenting with the view forming so far. The Framework Method, a way of analyzing interviews used since the 1980s, charts data in a similar grid: a spreadsheet with cases in rows and codes in columns⁶.

Keep could-not-say answers in the count. Three of eight people unable to name who approves a markdown is a finding, not missing data.

Check coverage halfway through and again a few days before the last interview. Log any held interviews first, then look at who is missing, not just how many. Five of six falls short if the sixth owned the project in question.

Most repairs are 5-minute calls about one question. Send questions nobody on your list can answer to the sponsor as open questions.

Write each theme in the interviewees' own words, from the matrix rather than memory. Give the count ("6 of 9 asked," not "most people"), the roles, one short quote, the dissent and what the theme can't support.

Treat the count as a description of the people you asked, not a poll. A 2013 paper on the Framework Method goes further, calling statements like "13 out of 20 participants said X" meaningless⁶, because interview samples aren't designed to represent a wider population. So write 6 of 9 asked, never 67%, and report an even split as a split.

Begin the report with four sentences: interviews held against those requested, questions you can report on, the ones you can't and who is missing, and what the reader won't find.

To skip the setup, open this template with Edit with AI and send your stakeholder list. River drafts the questions, targets and role guides. Once transcripts arrive, the interview synthesis tool counts how many people raised each theme.