River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Consulting Discovery Questions Template

Four documents and one sheet that score every discovery question against what it actually surfaced last time, not how good it sounds on paper.

Free download  ·  No account needed

Question Register  ·  how a status gets computed

Three statuses, and none of them is a guess

StatusTimes askedEngagementsHit rate
Core3 or more2 or more50% or more
Trialunder 3, orconfined to 1not decided yet
Retired3 or more2 or moreunder 50%

A hit needs a citation, not a good vibe. An answer only counts once it produced a finding cited downstream. An articulate answer nobody used again is a miss.

Two engagements, not one. Three perfect asks in a single interview are one client tested repeatedly, not a track record. The second engagement is what tells you a question travels.

Every row in the Question Register carries these three numbers, plus which point in the interview the question was asked at.

Search this and every result is a list: eight questions for stakeholder buy-in, a dozen for a change readiness call, thirty for a discovery interview sorted by role. All of it answers what should I ask, and none of it answers what actually matters by the third engagement: did asking it ever produce anything. A question that read brilliantly in one workshop three years ago and one nobody has used since sit on the same page in the same font weight, because nothing in the format tells the two apart.

So the unit here is the question, not the guide. Every question in the Question Register carries a status computed from three numbers: how many times asked, across how many engagements, and how many times the answer was cited in a finding. An answer nobody used again is a miss, not a hit, however articulate it sounded. Three asks across two engagements at half a hit rate earns Core, the same floor under half earns Retired, anything short stays Trial. Small wording changes shift what a question gets back, so wording stays fixed once live.

The bank behind this pack, run at an operations practice across four engagements, carries twenty scored questions: eleven Core, six Trial, three Retired. One of them, on who actually approves a fix, clears Core at a 77.8 percent hit rate overall. It still fails cold, at 33.3 percent, against 100 percent after a walkthrough, which the Sequencing Note tracks. A thin answer gets diagnosed against Gordon Willis's four stage model rather than re-asked louder. Run the interviews with the stakeholder interview programme pack, or price the hypotheses with the hypothesis and workplan pack.

Twenty questions scored across four engagements, and the two that look alike until you check what they produced

The Question Register, a Retired-versus-Core comparison, and the Sequencing Note.

Question Register

Six of twenty rows, for a fictional practice, Colston Reeve Advisory, across four engagements.

RefQuestionStatusAskedHitsRateEng.
DR-04Who actually has to approve the fixCore9777.8%4
DQ-01Which field do you not actually trustCore9888.9%4
PF-01What does a perfect handoff look likeRetired800.0%3
DQ-06What percentage of the time does it matchRetired4125.0%2
DQ-07What report did you last rebuild by handTrial22100.0%1
DR-03What decision were you unsure was yoursTrial2150.0%1

DQ-07 is a perfect two for two, and it still sits at Trial. One engagement is not a track record no matter how the two answers read. It needs a second client's transcript before it can move.

Across all twenty rows: eleven Core, six Trial, three Retired. Every number traces to a transcript or a citation, never to how promising a question sounded when it was drafted.

Two handoff questions, one word apart in spirit

Same boundary, the same four engagements available. Only one produced a citable answer.

PF-01PF-02
QuestionWhat does a perfect handoff look like?Walk me through the last time work came to you missing something.
StatusRetiredCore
Asked / hit8 / 08 / 6
Hit rate0.0%75.0%
Engagements34
Willis stage that failsJudgment: an ideal, not an instanceNone. Pulls a real event

PF-01 asks for a judgment about an ideal. Every one of its eight answers described how a handoff is supposed to work, and none of the eight was ever cited in a finding.

PF-02 asks for a specific instance instead. Six of eight answers named a real delivery and made it into a finding, because a dated event is checkable in a way an ideal is not.

Retired stays on the page with its failure named. Deleting it is how the next associate redrafts the same dead question in six months.

Sequencing Note: DR-04

Who actually has to approve the fix, not who's supposed to. Core overall, and not evenly.

PositionAskedHitsHit rate
Opening, cold3133.3%
After the walkthrough66100.0%
Aggregate9777.8%

Asked cold, this question gets the org-chart answer: whoever nominally owns approvals, whether or not they approved anything recently.

Asked after one real fix is already on the table, it gets a name. Verdict: ask after the walkthrough, avoid asking cold, on nine asks split three against six.

A verdict here needs three or more asks at the weaker position. Below that, the split is not reported at all.

What's in the pack

01

Question Register

Every question's real count: how many times it was asked, across how many engagements, and how many produced a finding someone actually cited back to it.

02

Question Bank by Domain and Role

Four domains, Core questions first in the order you would actually ask them, then Trial, then Retired with its failure mode named rather than deleted.

03

Probe Guidance

Gordon Willis's four stage model for why an answer comes back thin, comprehension, retrieval, judgment or response, and the different probe each failure needs.

04

Sequencing Note

Where in the interview a question performs best, read off its own position tagged history once two positions have real asks behind them.

05

Starting the Bank Cold

What an honest bank looks like at zero engagements, and which of your prior guides, transcripts already coded into themes, or synthesis writeups to send first.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send whatever guides, transcripts or a workshop record already exist, or download the Word documents and CSV sheet and start the register yourself.

  2. 2

    Score what you send against outcomes

    Every question actually asked in the material gets an ask logged against its engagement and position, and a hit only when the answer was cited in a finding, never when it merely sounded good.

  3. 3

    Let the count decide the status

    Three asks across two engagements at half a hit rate earns Core, the same floor under half earns Retired, and a question short of either stays Trial no matter how its one answer read.

  4. 4

    Carry the winners into the next guide

    The domain document keeps Core questions in asking order, Trial questions with what they need next, and Retired questions with their failure named, ready for the next stakeholder interview guide you build.

Frequently asked questions

Is this template free?

Yes. The zip is Word documents and one CSV sheet, no account and no card needed. Edit with AI is the optional half: the agent reads whatever guides or transcripts you send, scores each question, and builds the register from there. The rest of the library sits at the template library.

What format are the downloaded files?

Four Word (.docx) documents and one CSV (.csv) sheet, zipped together. Excel, Numbers or Google Sheets open the register straight off the download, and the four documents open in Word, Pages or Google Docs. Nothing needs converting first.

How is this different from a list of interview questions organized by topic?

A topic list tells you what to ask. This tells you how many times each question has been asked, by whom, and how often the answer became something you could cite. A question earns Core status from a count across two engagements, never from how good one answer sounded in the room. Turning a cited answer into a finding is the findings and recommendations pack.

What happens to a question that keeps coming back thin?

It gets diagnosed first, against the reason an answer comes back thin, comprehension, retrieval, judgment or response, detailed in a federal cognitive interviewing guide, and a matching probe replaces a louder repeat. Only once it clears three asks across two engagements with a hit rate under half does it retire, with the failure named rather than deleted.

Does it tell me when to ask a question, not just what to ask?

Yes, once there is enough history to say so. The Sequencing Note reports a position verdict only after a question has three or more asks at its weaker position. A question like DR-04 can clear Core overall at 77.8 percent and still carry a specific warning against asking it cold.

What does 'Edit with AI' actually do?

It signs you up, installs this pack as a private workspace, and puts the agent in front of an empty Question Register. Send it a prior guide, a transcript or a synthesis writeup, and it scores each question against what actually got cited, then reports what moved status and what to send next.

Does a question in the bank ever get tested at population scale?

Not by this pack directly. A Core question here reflects what already got asked and cited across engagements, not whether the underlying belief holds at population scale. The client survey and assessment pack is the next step for that: it turns a specific belief into a survey item, with a prediction staked before fielding. The two packs share the same evidence at two different scales.

Stop rewriting the same question bank

Take the Word documents and CSV sheet blank, or send River a prior guide or transcript and get the register scored before you draft next week's interviews.

Edit with AI