River
Y CombinatorBacked by Y Combinator
Log InStart Free
FREE TEMPLATE

Continuous Discovery Habits Template

Twenty weeks of interviews is twenty sets of notes unless something holds them together. This pack holds them in one log that never resets.

Edit with AIDownload Word and CSV

Free download  ·  No account needed

What you get

Pack of 7 files: 4 documents, 3 spreadsheets

Continuous Discovery Habits Template
  • How an Extract LandsDocument
  • Interview GuideDocument
  • Weekly CadenceDocument
  • Synthesis FormatDocument
  • Participant RegisterSpreadsheet
  • Insight LogSpreadsheet
  • Assumption TrackerSpreadsheet

A weekly interviewing habit does not fail on the interviews. It fails because each week's notes land somewhere separate, so by week twenty nobody is holding week one and the same finding arrives again as news. This pack has one Insight Log that never resets. Every extract either lands on a row that already exists or creates a new one, and because new rows are the exception, they are countable.

New rows divided by extracts, per five-week block, is the new-insight rate, and it is the only honest signal for whether more interviews on a question are worth running. It falls for two different reasons though, and the rate alone cannot separate them. Nielsen's diminishing-returns curve carries the same caveat: the formula holds only for comparable users, and genuinely different groups have to be covered separately. So every participant carries a recruitment source.

In the worked example that rate fell from 48 percent to 14 across fifteen weeks and looked finished. Then the source column showed 30 of 34 interviews had reached customers who were already succeeding. Four from churned accounts and a paid panel produced 17 new rows from 29 extracts, on the same question, in week seventeen. Built for a team whose standard is at least weekly contact with customers and whose week keeps not having it in. It replaces the study-shaped practice that research synthesis serves well and a weekly cadence outgrows.

Twenty weeks, one log, and the rate that comes out of it

The block rates, the recruitment sources behind them, and the assumptions the interviews never touched.

Continuous Discovery Habits Template
RiverNew-insight rate

New-insight rate

Illustrative, for a fictional physiotherapy practice-management product called Halyard. New rows created divided by extracts produced, per five-week block. 34 interviews against a plan of 40.

WeeksInterviewsExtractsNew rowsRateSourcesPopulations
1 to 59 of 10713448%21
6 to 109 of 10661523%21
11 to 158 of 1057814%21
16 to 208 of 10532038%43
Twenty weeks3424777-43

Weeks 16 to 20, split by who was recruited

Recruited fromInterviewsExtractsNew rowsRate
The usual in-app prompt424313%
Churned accounts and a paid panel4291759%

Read the first three rows alone and the conclusion is obvious: the question is answered, go and build. The team proposed exactly that at week fifteen. The populations column is the tell and it had been reading 1 since week five.

RiverParticipant Register

Participant Register

Every participant carries where they came from and what population that channel actually reaches. Two sources reaching the same people is one source.

Recruitment sourceInterviewsSharePopulation it reaches
In-app prompt2265%Active users already succeeding
Customer success referral824%The same population
Churned accounts39%Practices that left inside 90 days
Paid panel13%Non-customers evaluating a change

How many sources sit behind each of the 77 insight rows

Sources behind the rowRowsShareNote
Three or more912%Safe to act on
Exactly two2127%Usually the same population twice
Exactly one4761%41 of them the in-app prompt

Sort the log by source count rather than evidence count once a month. The largest row in this one has 23 extracts from 19 participants and three sources, which is a finding. The next one down has 19 extracts and one source, which is a property of the in-app prompt.

RiverAssumption Tracker

Assumption Tracker

One number per assumption: how many interviews have produced evidence bearing on it, either way. Not whether anybody believes it.

AssumptionBy week 15By week 20Status
Practices accept a two-week window with old notes read-only0 of 263 of 34Contradicted
The front desk, not the clinician, decides whether it stays3 of 264 of 34Under-evidenced
Rejected insurance claims are the largest admin cost14 of 2617 of 34Well evidenced
Three or more clinicians means cross-site scheduling11 of 2611 of 34One population only
Clinicians will use a tablet at the bedside5 of 266 of 34Contradicted
Below $34 per clinician a month, price stops deciding2 of 262 of 34Untested

The comparison is the finding rather than either number. Twenty-six interviews in, the assumption gating the largest item on the roadmap had none behind it and the assumption nobody in the building disputed had fourteen. Three churned-account interviews then settled the first one, against it.

Edit with AI

What is in the pack

01

Insight Log

One row per problem, never per study. Extracts land on rows that already exist, carrying an evidence count, a participant count, and how many recruitment sources sit behind them.

02

The new-insight rate

New rows divided by extracts, per five-week block. The one number that says whether more interviews on this question are still worth running, rather than whether they felt productive.

03

Participant Register

Recruitment source on every row, plus a sentence naming the population that channel actually reaches, because two sources often reach the same people.

04

Assumption Tracker

Each belief the roadmap rests on, carrying how many interviews have touched it either way, and the specific interview that would settle it.

05

Interview Guide

Openings that get a specific recent instance rather than a process description, the four follow-ups that find workarounds, and the four questions that produce rows which never resolve.

06

Weekly Cadence and Synthesis Format

A week that survives a release, a five-part Friday synthesis that fits one screen, and a restart procedure that goes back into the same log rather than a clean one.

How it works

  1. 1

    Send what already exists

    Old transcripts, the findings deck nobody reopened, past reports, the list of people you meant to talk to. It becomes the log before the first new interview.

  2. 2

    Interview twice a week

    Two conversations, recruited against the register rather than the recruiting tool, each one recording where the participant came from and who that channel reaches.

  3. 3

    Land the extracts

    Within the day. Each extract joins a row that already exists, or creates one with a sentence saying why nothing matched. Creating is the exception and it gets counted.

  4. 4

    Read the rate, then the sources

    Every five weeks. A falling rate means the question closed or the channel ran out, and only the source count tells you which one happened.

Frequently asked questions

We already do weekly interviews. What does this add?

A number for whether they are still teaching you anything. Weekly attendance is a vanity metric: ten weeks of the same person type is a well-attended programme producing nothing, and it looks green on every dashboard. The rate and the source count are what make that visible from inside.

What counts as a new insight rather than an existing one?

Match on the underlying problem, never the wording. Printing the schedule every morning and not seeing tomorrow on a phone are one row. The words stay on the extract and the row is the problem, which is what stops a log growing four thin rows describing one thing.

Is a falling new-insight rate a good sign?

Only alongside three or more recruitment sources reaching genuinely different populations. Below that, the honest reading is that the channel is exhausted rather than the question answered. In the worked example the rate went from 14 percent back to 59 on the same question, from a different channel.

What happens when the habit stops for a month?

It will, usually around a release. The restart goes into the same log: read the ten rows with the oldest last-touched date, pick the two that still matter, and book against those. Opening a fresh register is what destroys the accumulation, and it feels productive immediately.

Does this decide what we build?

No, deliberately. Insights become candidate solutions and bets somewhere else, and the ranking of what ships belongs in the prioritization framework. A discovery habit that starts recommending stops producing evidence within about a month, because the interviews get run to confirm a decision already taken. Testing one of those bets before it is built is the concept test pack.

How does this relate to our other feedback sources?

They answer different questions and the log holds insights, not requests. Tickets, reviews and survey comments are covered by the voice of customer review, and a feature request board belongs in customer feedback triage. Neither reaches the people who never became customers.

Who should be in the interviews?

Two people from the team, one asking and one writing extracts live. Never three, because a third turns it into a meeting and the participant starts performing. The person building the thing should be one of the two at least every other week.

Sources

  1. Nielsen Norman GroupWhy You Only Need to Test with 5 Users (opens in a new tab)
  2. Product TalkProduct Discovery Basics: Everything You Need to Know (opens in a new tab)

Written by

Chandler Supple

Co-founder & CEO, River

Chandler has spent the last six years working in AI, building tools for millions of users and Fortune 100 companies. In 2025, he co-founded River to build a new kind of operating system for AI agents.

LinkedInGitHubX

Reviewed by

Brian Joseff, Co-founder, River

LinkedInX

Related

Jobs to be done and persona pack

Turns an accumulated log into job statements and personas that carry the interviews behind them.

Usability test report

The other half of a discovery week, where severity comes from task failure rather than researcher judgement.

Win-loss and churn input review

Reaches the population an in-app prompt cannot: the deals that were lost and the accounts that left.

More free templates

Concept Testing Plan Template

Free concept testing plan template that freezes the kill criteria before the first session and scores what participants did rather than what they said.

Jobs to Be Done Interview Analysis

Free JTBD template turning transcripts into job statements that each carry the interview count behind them, plus the anti-persona that lets you say no.

Research Repository Structure Template

Free research repository template indexed by question rather than by study, with an expiry on every answer tied to a named change instead of a date.

Accessibility Audit Remediation Plan

Free remediation plan that regroups your axe or WAVE export by the component that rendered each selector, then ranks the fixes by what they actually block.

Find out what your last twenty interviews actually added

Send whatever notes exist, however old and however messy. They become the log, and the first number back is how much of it is one channel.

Start the log