How Many Customer Interviews Is Enough — Write the Stopping Rule Before You Count

1. Segment — one job, one context, one buyer

A segment is a group that shares the same recent job and constraint — not “SMBs,” not “people who might like this.” Guest, Bunce and Johnson caution that their 12-interview finding fits relatively homogeneous groups. Mix two jobs and you are counting coverage you do not have.

2. Saturation — a full batch adds no new important code

Saturation here means no new important need or pain codes in a complete batch. It does not mean people liked the idea. It does not mean they will pay. Hennink and Kaiser are explicit: their 9–17 range is not a generic rule, and outliers (multi-country work, meta-themes, code-meaning saturation) needed larger samples.

3. Batch — about 5–6 talks, then you recode

A batch is a small set you finish before you decide whether to continue. Guest, Bunce and Johnson saw basic elements of metathemes as early as 6 interviews. Treat about 5–6 as a starting batch size derived from that early-theme finding and from the practical need to recode, not as a law.

4. Stopping rule — six fields, written before call 1

Segment, batch size, minimum N, saturation check, cap, and the named action. If any field is blank, you do not have a rule. A number invented after a warm week is a recap.

5. Coverage vs validation

Coverage asks: have we heard most of the needs and pains in this segment? Validation asks: will they pay, use, or commit? Interview count measures the first. Deposits, checkouts, smoke-test clicks, and scarce commitments measure the second. Do not convert one into the other.

Mom Test, discovery, solution interview — how to ask, not how many

Those pages are conversations: last-week behavior with no pitch, the discovery process, or a scripted artifact show after the problem is evidenced. This page is how many talks in one segment and when to stop. A good question is not a saturation check. A completed walkthrough is not coverage.

Product-market fit survey — a usage percent, not saturation

That page’s signal is the percent of recent real users who would be very disappointed if the product disappeared. This page’s signal is whether a batch in one segment still adds new need codes. Saturation is not a Sean Ellis percent. A very-disappointed score is not interview coverage.

WTP, smoke, fake-door, landing page — behavior and money after coverage

Those pages measure reach, clicks, deposits, or checkouts. They usually sit after interviews saturate — or instead of more interviews when the expensive unknown is already money or reach. A waitlist is not a new need code. A deposit is not saturation.

Earlyvangelist and design partner — who, and scarce commitment

Those pages filter a named person (Blank’s five characteristics) or a scarce co-dev commitment. This page counts coverage inside a segment you already named. Matching the five is not saturation. A signed partner is not a batch of codes.

Kill, pivot, assumption, decision — contracts that can later read this rule

Those pages write Stop, rewrite a dimension, name one claim, or read a week of evidence you already have. This page supplies one input those contracts can later read: a dated coverage check in one segment. It is not a Stop / Pivot / Continue sitting.

Yibud scores — not a saturation measure

The validation-score and score-calculator pages explain Startup MRI numbers from a deterministic engine on a one-line idea plus five questions. Those scores are flashlights on assumptions. They are not interview saturation and not a Guest-or-Hennink count. Do not read a mid-band score as “enough interviews,” and do not read 12 talks as a Yibud GO.

The stopping rule

Segment: [one job + one context + one buyer — not “SMBs”]. Batch size: [about 5–6 starting size]. Minimum N: [at least one full batch]. Saturation check: a full batch adds no new important need/pain code. Cap: [~20–30 in one segment → split; Griffin & Hauser hypothesis, one category]. Action when the rule fires: [stop this segment / split it / run a named money or behavior test]. Codes you will log: [need and pain phrases in their words].

If any bracket still says “customers,” “enough,” or “later,” the contract is not written. Interviews from two jobs do not share a count. Compliments are not codes.

What does not count as the signal

Not this signal: a Yibud score, a compliment, “I’d use that,” a Sean Ellis very-disappointed percent, a waitlist, a smoke or fake-door click, a deposit or checkout, matching Blank’s five, a signed design partner, a completed solution-interview walkthrough, Nielsen’s 5 usability users. Optional later: those instruments have their own pages.

You may still take notes on praise or money. You may not let them fire this stopping rule. New important need/pain codes in a completed batch fire this rule — or the absence of them does.

Example starting rule (template — the counts are the research defaults, labeled)

In [named segment], interview in batches of 6. After each batch, list new need/pain codes. Minimum: 6. Stop when a full batch adds no new important code (expect roughly 9–17 in a homogeneous segment — Hennink & Kaiser 2022, not a law). If you pass ~20–30 and still hear new needs, split the segment (Griffin & Hauser 1993 hypothesis, one category, per segment). Action: stop this segment and run a named WTP or smoke test.

9–17, 12, and 20–30 are from the studies named on this page, labeled as findings or hypotheses. They are not Yibud results and not a claim about what usually happens in your category. This page invents no other N.

The stopping rule, written the same sitting

Segment: as above. Batch size: 6. Minimum: one full batch. Saturation check: a completed batch adds no new important need or pain code. Cap: 24 in this segment — if new important codes continue, split (Griffin and Hauser’s 20–30 hypothesis, adopted as a cap, labeled as theirs). Action if the rule fires: stop interviewing this segment and run a named deposit test on the two recurring pains. Compliments do not count as codes.

What a stop looks like after three batches

Illustrative close: batch 1 (talks 1–6) adds many codes — Friday-night chasing, clients who will not open a portal, shame about late invoices. Batch 2 (7–12) adds two codes. Batch 3 (13–18) adds none that the founder marked important. That is a coverage stop you already named — move to money evidence. It is not a reason to book twenty more “because people were warm.” Eighteen talks in this fiction is not Guest’s 12 and not a claim about your week.

What must not happen in the window

Do not fold agency bookkeepers and in-house controllers into the same count. Do not treat “I’d use that” as a new need code. Do not borrow Nielsen’s 5 and stop after five compliments. Do not treat a Yibud score as saturation. Do not keep going past the cap because the stories rhyme — that is a split, or a new page with a reason. If you cannot find one batch in the named segment, you may have a reach problem — discovery or smoke, not a fake pass.

Misusing Nielsen’s “5 users” as a discovery sample

Nielsen (2000) wrote about usability testing of an existing design: about 85 percent of usability problems with 5 users, and several small rounds beat one big one. That is not a discovery-interview count. Five compliments about a slide are not saturation.

Counting interviews across mixed segments

Twelve talks with three different jobs is not Guest’s 12. Coverage is per segment. Griffin and Hauser: even the 30 customers in that study did not capture every need; their model estimated those 30 customers accounted for 89.8 percent of all needs, because low-probability needs were still missing. A new segment beats a 40th interview in the old one.

Treating saturation as validation

No new themes means you have probably heard the needs in that group. It does not mean they will pay, click, or stay. After the rule fires, move to a named money or behavior test. Do not announce product-market fit.

Stopping on compliments — or never writing the action

“I’d use that” is politeness. It is not a new code and not a reason to stop early. If you did not name the action before call 1, every warm talk becomes “qualitative insight.” Write the fail line on the same page as the segment.

Treating 12 or 20–30 as a law or a trophy

Guest, Bunce and Johnson caution that 12 fits homogeneous groups and focused aims. Hennink and Kaiser say 9–17 is not a generic rule. Griffin and Hauser’s 20–30 is a hypothesis from one product category, per segment. Adopt the starting rule or rewrite it before the first call. A band invented after a messy week is a recap.