River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Stakeholder Interview Guide Template

Five documents and three sheets built around one number: how many of the people who hold a view on each question have actually answered it.

Free download  ·  No account needed

Interview Guide by Role  ·  Warehouse supervisor  ·  v3

Forty minutes, six blocks, in this order

BlockMinutesWhat it is forRefs
Framing2Who commissioned this, what it covers, that nothing is attributed by name, that you can stop at any point.
Their job5What they actually do, not what the role description says. Establishes which questions apply.
Walkthrough12One recent real instance, start to finish. Always before the programme questions.
Programme questions16The numbered set, same references in every guide, so the answers are comparable.QC-01, QC-03, QC-07
Not applicable0Marked in the guide before fielding, which lowers the target rather than the count.QC-02, QC-06
Who else3Who else should be asked this, and who disagrees with you. The highest yield minute in the hour.

The walkthrough comes first for a reason. Ask a general question cold and you get the documented process. Ask it after twelve minutes on last Tuesday's order and you get what happened on Tuesday.

The references are the point of the margin column. Every guide carries the same numbered questions, so an answer given by a warehouse supervisor and an answer given by a store manager land in the same row of the matrix and can be counted together.

Version 3 because QC-08 was added in week two. Everyone interviewed on v1 and v2 gets a five minute follow up rather than being counted as covered.

Search this and you get question banks. Four topic areas, a discussion guide, a note-taking canvas, thirty questions to adapt by role. All useful, and none of it is the part that fails. What fails is arriving at the write-up with twenty conversations and no way to say how many people a finding rests on. Six people said the handover is broken is a claim about six people. Six of nine is a claim about the organisation.

So the unit here is the question, not the interview. The coverage matrix holds one row per question with the roles that hold a view, the target, how many were actually asked, how many agreed, how many dissented, how many could not say, and who was never asked at all. Could not say stays in the denominator: three of eight buyers unable to name who approves a markdown is the finding, and it is the answer that gets dropped first.

Targets get set before the first interview, which is the ordering that matters. OMB's standards for statistical surveys require a nonresponse bias analysis whenever a response rate lands under 80 percent, and require it planned in advance where the expected rate already is. None of that arithmetic transfers to fourteen interviews. The discipline does: set the target first, then notice the shortfall in writing while another conversation can still be booked. Open the pack in River once you have indexed the client's data room, or take the Word documents and CSV sheets blank.

The matrix that decides which questions you are allowed to report

Every count in the Theme Register comes from the matrix. Nothing in it is recounted from the transcripts.

Question Coverage Matrix

Illustrative rows for a fictional retail group, Kestrel, in week four of six.

RefQuestionTargetAskedAgreedDissentCannot sayVerdict
QC-01Where does an order stop moving99720Analysable
QC-02Who decides a markdown88413With a caveat
QC-03What happened to the 2024 project64130Not analysable
QC-04How long a stock query takes77700Analysable
QC-06Is consolidation reversible31001Not analysable
QC-07Which system do you key into99360Analysable

QC-03 is short by two and both of them matter. Four of six asked, three of the four contradict the official account, and the two never asked include the director who owned the project. Nothing from this question goes in the report until that interview happens.

QC-06 has one holder who could not say. One of three asked, and that one did not know. This does not become a cautious finding. It leaves the themes entirely and goes to the sponsor as an open question.

QC-07 is the reason the programme was run. Six of nine describe something different from the documented process, and that disagreement is only reportable because the question was put to all nine.

Cannot say sits in the denominator. QC-02 is eight of eight asked, and the three who did not know are the finding rather than a gap.

Theme Register

A theme cannot be written here without a count and a denominator.

RefThemeOfDissentStrengthWhat it cannot support
T-01Orders go into the legacy screen because the new one rejects part quantities6 of 93StrongHow often part quantities occur. Nobody was asked and no system data seen
T-02Nobody can name who signs off a markdown below thirty percent4 of 81ModerateThat no approval process exists. Three said they could not say, which is not the same
T-03The 2024 project is described differently at director level and on the floor3 of 41Not analysableAnything at all yet. The split is between four people, not between two levels
T-04A stock query takes two days and everybody says two days7 of 70StrongTwo days as a measured cycle time. It is what seven people expect
T-06Directors split evenly on whether twelve months means fixing or replacing4 of 52ModerateA recommended direction. An even split is reported as a split, never as a majority

The last column is the one that earns its place. T-04 is unanimous and it still cannot support a cycle time, because nobody was asked for one. Writing that down is what stops two days appearing in a deck three weeks later as though somebody had measured it.

Dissent is a column, not an omission. T-01 sits at six of nine and names the three who use the new screen and have never hit the limit. That is what survives an executive saying nobody has ever told them this.

T-03 stays in the register at not analysable rather than being deleted. It is the reason to book the director.

Interview Schedule and Tracker

A decline stays in the sheet. It changes the target rather than leaving a blank.

RefRoleGuideHeldConsentLoggedStatus
I-01Store managerStore v209-08On recordingYesComplete
I-03Transport plannerPlanner v109-10In writing, not recordedYesComplete
I-05Supply chain directorDirector v1Scheduled, blocking
I-06Finance partnerDirector v1Requested twice
I-07IT managerIT v109-12On recordingNoTranscript held
I-08Customer service leadStore v2Declined
I-10Warehouse supervisorWarehouse v309-18On recordingYesComplete

I-07 is why this column exists. The transcript arrived and was never posted against the questions, so every count in the matrix understates by one. A coverage check run on those numbers sends you chasing somebody you already interviewed.

I-03 declined recording, which is fine and changes one thing. Every quote from that conversation is marked as a paraphrase in the register and in the report, because a paraphrase and a verbatim are not interchangeable in front of a client.

I-10 was added in week three. Three earlier interviews named him as the person orders route around. The empty slots left at the end of the schedule exist for exactly that.

I-08 declined and gave a reason worth keeping: she was interviewed for the last review and nothing came of it.

What's in the pack

01

Question Coverage Matrix

One row per question, a target set before fielding, and columns for asked, agreed, dissented, could not say and never asked.

02

Theme Register

No theme without a count and a denominator, the dissent named, and a column for what the theme cannot support.

03

Interview Schedule and Tracker

Guide version per interview, consent recorded and how, and a logged-to-matrix flag so you know whether the counts are current.

04

Interview Guide by Role

Six blocks in a fixed order, with the walkthrough before the numbered questions and the same references carried in every role's guide.

05

Asking a Question Enough Times

How the target gets set, the four coverage verdicts, and why a shortfall of two can be worse than a shortfall of five.

06

Consent and Confidentiality Note

Built to the shape federal informed consent rules require: key information first, and nothing asking the participant to waive anything.

07

Programme Conventions

Eight decisions to settle before interview one, including what counts as an interview and who ever sees the interview level record.

08

Synthesis Format

The seven parts every theme takes, and the four sentence paragraph on coverage that goes above all of them.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send it the stakeholder list, or download the Word documents and CSV sheets and run the programme yourself.

  2. 2

    Set a target per question

    Eight to twelve numbered questions, each with the roles that hold a view and how many of them you intend to reach. Anything only one person can answer becomes a briefing instead.

  3. 3

    Log against questions, not into files

    Each transcript gets posted question by question, so the count moves as the interview happens rather than being reconstructed from write-ups at the end.

  4. 4

    Check coverage while you can still fix it

    Run the check at the halfway point. It names who is missing, ranks the repair by what the next conversation buys, and flags the questions one person alone can settle.

Frequently asked questions

Is this template free?

Yes. The zip is Word documents and CSV sheets, no account and no card. Edit with AI is the optional half: the agent writes the question set, sets the targets, builds the role guides and logs each transcript against the matrix. Other packs sit in the template library.

What format are the downloaded files?

Word (.docx) for the five documents and CSV (.csv) for the three sheets, zipped together. Excel, Numbers and Google Sheets open the matrix, the register and the tracker straight off the download. Nothing needs converting first.

How is this different from a list of interview questions?

A question bank tells you what to ask, and a bank that scores each question against what it produced last time is where the set should come from. This tells you how many times each was asked this round, and who is still missing. Deciding which hypotheses justify an interview round at all is the hypothesis and workplan pack.

How many people is enough for one question?

It depends on who is missing rather than on the count. The pack sets a target per question before fielding, then grades each one as analysable, analysable with a caveat, not analysable, or withdrawn. Five of six is not analysable when the missing person owned the thing you are asking about.

Does it work from recordings and transcripts?

Yes, and from notes where somebody declined recording, in which case every quote is marked as a paraphrase. If you want a single conversation turned into quotes with their timecodes rather than counted across a programme, working up one transcript does that. A group session is a different write-up: a workshop record built from the board photographs keeps what the room parked.

Will it name who said what?

No. The matrix and the register carry roles and interview references, never names, and it flags any quote whose content would identify the speaker even without one. That is the promise the consent note makes to participants, so the artifacts have to keep it. Once fielding is done, clearing each verbatim against the org chart decides which quotes reach the deck.

What does 'Edit with AI' actually do?

It signs you up, installs this pack as a private workspace of your own, and puts the agent in front of an empty coverage matrix. Send the stakeholder list and what the engagement has to settle, and the question set, the targets and the role guides come back before you book anybody.

Know what twenty interviews establish

Take the Word documents and CSV sheets blank, or open this exact pack in River and send it the stakeholder list before you book anybody.

Edit with AI