River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Consulting Client Survey Template

Three documents and three sheets that stake a prediction and a segment cut before the survey opens, then compute what the population actually confirms.

Free download  ·  No account needed

Verdict Rule  ·  computed once the field closes, never before

Three verdicts, and only one needs no floor

VerdictStaked before sendClears the floorObserved result
ConfirmedYesYesMatches the prediction
ContradictedYesYesMisses the prediction
InconclusiveNo, orNonot evaluated

A cut with no stake cannot confirm or contradict anything, however the rate reads. Both real verdicts require a prediction written down before the send, aimed at that exact segment.

A cut under the floor stays Inconclusive at any rate. A depot in the worked example agreed at 53.3%, which would have confirmed a company-wide prediction, and it ships Inconclusive because its 15 respondents sit under this pack's own floor.

Every row in the Segmented Results sheet carries its own respondent count, so the check is arithmetic, never a judgment call.

An interview round is fast and the sample is always small: nine people, eleven, fourteen. Whatever a handful of them said tends to get written up as what the organisation believes, and the write-up rarely says how many of the eleven actually said it. Turning a set of recordings into thematic findings already tracks the count behind each theme. What nothing in the category does next is check whether that count, real as it is, generalises to the hundreds of people the client actually has to manage.

This pack turns a theme into a survey item only once it clears three tests: decision-relevant, answerable without guessing, and either population-wide or already split by segment at interview. Every surviving item carries a Stated Prediction and a Segment Cut, both dated before the survey opens and locked once it does. ICH's own statistical principles for confirmatory trials treat an anticipated subgroup comparison, written into the plan first, as confirmatory, and one found only afterward as exploratory and unlikely on its own to support a conclusion.

In the worked illustration, dispatchers at one depot said, five of six, that they skip the delivery-window check before promising a time; at another depot, one of three said the same. The survey staked that gap and confirmed it, 78.8% agreement at the first depot against 37.5% at the second. Supervisors told the opposite story: five of six said, at interview, that route changes skip the fuel-budget check, and 55 surveyed supervisors confirmed only 40.9%. Preregistering a prediction before the data exist is what makes a Contradicted result trustworthy rather than convenient.

Four items staked before the survey opened, and what came back once it closed

The Question Bank, Segmented Results and one depot's near-miss, from the fictional worked example.

Question Bank mapped to hypotheses

Four of the pack's items, for a fictional logistics client, Osgrove Logistics.

RefTypeInterview basisStated predictionSegment cut
SQ-01Frequency3 of 11 drivers said 3+ times a week, unpromptedAt least 25% company-wide, 3+ times weeklyNone
SQ-02Agreement4 of 11 drivers, no site patternAt least 40% company-wideNone
SQ-04Agreement7 of 11 dispatchers: Northgate 5/6, Riverside 1/3At least 55% company-wide; Northgate materially higher than RiversideNorthgate vs Riverside (staked)
SQ-07Agreement5 of 6 supervisors, company-wideAt least 60% company-wideNone

The prediction is dated the day the survey opens, never after. SQ-04 is the only row with a segment cut staked, because Fenchurch's interview base, one dispatcher of two, was too thin to call a direction.

Six more items sit in the full sheet. A prediction with no interview basis behind it does not ship.

Segmented Results

What came back once the field closed, fifteen days after opening.

RefSegmentNRateClears floorVerdict
SQ-04Dispatchers, company-wide7259.7%YesConfirmed
SQ-04Northgate3378.8%YesConfirmed
SQ-04Riverside2437.5%YesConfirmed
SQ-04Fenchurch1553.3%NoInconclusive
SQ-07Supervisors, company-wide4440.9%YesContradicted
SQ-02Drivers, company-wide8434.5%YesContradicted

Northgate and Riverside confirm the same row two different ways. Their 78.8% and 37.5% were staked as a pair, Northgate materially higher than Riverside, and the 41 point gap between them clears easily.

Fenchurch sits between them at 53.3%, and gets no verdict. Fifteen respondents do not clear this pack's own floor, and no prediction was staked on that depot alone.

SQ-04, depot by depot

Dispatchers only: do you check the system's predicted delivery window before promising a customer a time.

DepotResponded / populationAgreedRateStakedVerdict
Northgate33 / 402678.8%Yes, pairedConfirmed
Riverside24 / 34937.5%Yes, pairedConfirmed
Fenchurch15 / 22853.3%NoInconclusive

Fenchurch's own rate, 53.3%, would have confirmed the 55% company-wide line if it were the only number that mattered. It is not reported that way, on two independent grounds: its 15 respondents miss this pack's own 20 floor, and its thin interview base never had a direction staked on it to confirm.

Extending Fenchurch's field window by a week was flagged while there was still time. Extending it after the close, once the number already looks good, is exactly the after-the-fact call the Stated Prediction exists to rule out.

What's in the pack

01

Survey Design

Which interview themes clear the three tests, decision-relevant, testable, and either population-wide or already split by segment, before any of them becomes a question.

02

Question Bank Mapped to Hypotheses

Every item's interview basis, its Stated Prediction and Segment Cut, and the date both were staked, the same discipline behind a discovery question's own hit-rate history.

03

Distribution Plan

Channel, reminder cadence, anonymity safeguards, and the rule for extending a segment's outreach before its close date rather than after.

04

Response Tracker

Invitation, reminder and response dates per respondent, tracked by role and site only, never by name.

05

Segmented Results

Respondent count, observed rate, the 20-respondent floor check, and the verdict against the prediction staked on that exact cut.

06

Results Narrative

Company-wide and by-segment findings written up as Confirmed, Contradicted or Inconclusive, in that order of how the count actually landed.

07

Response Reminder Sweep

A daily fielding sweep that drafts the next reminder in its locked wording and flags a segment tracking under the floor.

How to use it

  1. 1

    Open in River, or take it blank

    Open the pack in River and send the themes from a completed interview round and whatever population data exists, or download the documents and sheets and start the bank yourself.

  2. 2

    Turn themes into items

    Each surviving theme clears three tests, decision-relevant, answerable without guessing, and population-wide or already split by segment, then gets a Stated Prediction and Segment Cut dated before the survey opens.

  3. 3

    Field it and watch the floor

    The sweep drafts each day's reminder in its locked wording and flags any segment projected to land under twenty respondents while there is still runway to extend it.

  4. 4

    Compute the verdict, not the vibe

    Once the field closes, every staked cut gets Confirmed, Contradicted or Inconclusive from its own respondent count against its own prediction, written into the Results Narrative.

Frequently asked questions

Is this template free?

Yes. The zip is three Word documents and three CSV sheets, no account and no card needed. Edit with AI is the optional half: send the interview themes and the agent drafts the survey items, each with a prediction and a segment cut, before anything is fielded. The rest of the library sits at the template library.

What format are the downloaded files?

Three Word (.docx) documents and three CSV (.csv) sheets, zipped together. Excel, Numbers or Google Sheets open the Question Bank, Response Tracker and Segmented Results straight off the download, and the three documents open in Word, Pages or Google Docs. Nothing needs converting first.

How is this different from a client satisfaction survey?

A satisfaction survey asks how people feel about the engagement in general. This one starts from a specific interview theme, like five of six dispatchers at one depot skipping a delivery-window check, and asks whether that holds at population scale. Every item traces to a named theme from the interview transcripts rather than to a generic scale.

What has to be written down before the survey opens?

A Stated Prediction and a Segment Cut, both dated, for every item. The prediction names the threshold or direction the interview stage suggested, and the cut names which population split, if any, the item is tested against. A theme that is a data question rather than a belief probably belongs in the hypothesis and workplan pack instead.

What happens when a segment doesn't clear twenty respondents?

It stays Inconclusive regardless of what the rate reads. In the worked example, one depot's dispatchers agreed at 53.3%, a rate that would have confirmed the company-wide prediction, but its 15 respondents sit under this pack's own floor of twenty. Federal disclosure-limitation practice suppresses a small cell too, though the published threshold rule sets that minimum at three to five rather than twenty.

Does a Contradicted result mean the interviews were wrong?

No, it means the theme did not generalise past whoever raised it. Five of six supervisors said at interview that route changes skip a fuel-budget check, and only 40.9% of 44 surveyed supervisors agreed. The interview theme was real. The population majority the write-up would otherwise have implied was not.

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 Bank. Send it the interview themes and whatever population data you have, and it drafts each item's prediction and segment cut, then reports every verdict once the tracker shows the field closed.

Stop reporting an interview theme as if it already generalises

Take the Word documents and CSV sheets blank, or send River the interview themes and the client's population data and get the Question Bank staked before the survey opens.

Edit with AI