Five-Customer Test — The Product Playbook

It's Friday, the last day of a five-day Design Sprint. The team spent Monday through Wednesday arguing toward one idea, Thursday building a fake front of it, a screen that looks real but has nothing built behind it, and now a real customer is about to sit down in front of the thing and try to use it. This is the day the whole week was built around. Every map, every sketch, every vote was just the road to this room.

And here's the part that stops people cold. You don't need a hundred customers. You don't need fifty. You need five.

Five one-on-one interviews, one at a time, each about an hour, with the prototype in front of them and the whole team watching. By the end of the day you'll know whether Monday's bet survived contact with a real person. Monday's bet is the one idea the team committed to on day one, the assumption this whole test exists to confirm or kill. Five customers. One day. That's the test.

Why five

Five sounds too small to trust. It isn't, and the reason is a number you can lean on.

Jakob Nielsen of the Nielsen Norman Group put it in writing back in 2000. Watch about five real people try to use a thing, and you'll surface roughly 85% of the spots where people get stuck or confused. Not 85% of every problem that exists. About 85% of the obvious ones, the ones you'd actually act on. After the fifth person, you start watching the same problems show up again, and you stop learning much that's new.

So the math runs against your instinct. A bigger test doesn't buy you a better answer. It buys you the same answer, slower and more expensive. Nielsen's own point is that you're better off running three small tests of five than one giant test of fifteen, because the wins come from fixing what you found and testing again, not from piling up more people on the first pass.

One honest caveat, because it matters. That 85% is a rule of thumb for watching people use something, where you read behavior and follow up live. It isn't a statistic for a survey or a numbers study, and it doesn't mean five people speak for your entire market. It means five is enough to learn what's broken on this prototype, this week. That's all you came for.

Knapp borrowed Nielsen's number for the sprint for a second reason that's almost too convenient. Five one-hour conversations, with breaks, happen to fit in a single day. The research says five is enough, and the clock says five is what fits. So Friday is five.

The shape of the hour

Knapp, Zeratsky, and Kowitz run Friday's interview as a five-act structure, a script Knapp credits to Michael Margolis, the research partner at GV (Google Ventures). Five acts, in order, each doing one job. Walk it the way you'd actually run it.

1. A friendly welcome. You start warm, not with the prototype. Thank them for coming, keep it casual, and tell them the honest truth. There are no right answers, you're testing the thing and not them, and the only way they help you is by being straight. Mention the team is watching, so it isn't a surprise. The whole job of this act is to get a nervous stranger to relax, because a tense person performs and a relaxed person tells you the truth.

2. Easy context questions. Before the prototype comes out, ask about their life and their work. Not your idea. Their world. What does a normal day look like, what tools are they using now, where does the thing you're testing even fit. This warms them up, and it tells you who you're actually talking to, so that when they react to the prototype you understand the person reacting.

3. Introduce the prototype. Now you bring it out, and you set expectations first. Tell them some things might not work, because it's early. And tell them the small white lie that does a lot of work. You didn't design this yourself. Say something like "someone else on the team put this together," so it isn't yours to defend. The moment they think it's your baby, they soften every word to spare your feelings, and polite is the one thing you can't use. Once they don't own it, they'll tell you a screen is confusing instead of telling you it's nice.

4. Tasks, and a nudge, not a lead. This is the act everything else exists to protect. You give them a real task, something they'd actually do, and you ask them to think out loud while they do it. Then you sit on your hands. You do not demo it. You do not show them where to click. You do not rescue them when they're lost, because the spot where they get lost is the most valuable thing in the room. If they stall, you nudge with a question ("what are you trying to do right now?" / "what would you expect that to do?"), never a direction. The second you lead them, the test is over and you just don't know it yet.

5. A quick debrief. When they've worked through the tasks, you zoom back out. What stood out, what was confusing, how does this compare to what they do now, would this fit into their actual day. You're collecting their overall read and giving them a clean place to land.

That's the hour. Welcome, context, introduce, tasks, debrief. Repeat it five times.

Watch what they do, not what they say

If you take one rule off this page, take this one. Read behavior, not opinions.

A customer will tell you they love it. They'll tell you they'd use it, they'd buy it, it's exactly what they needed. And they'll say all of it while their hands are stuck on a screen they can't figure out. The words are politeness. The hands are the truth. When the two disagree, believe the hands every time.

This is the whole discipline behind a good interview, and it's bigger than the sprint. The deep version, how to ask so people can't flatter you, lives on the Talking to Customers page, and the cleanest single book on it is The Mom Test by Rob Fitzpatrick. His three rules carry straight into Friday's room: talk about their life instead of your idea, ask about what they've actually done instead of what they'd hypothetically do, and talk less so you can listen more. Fitzpatrick's own line is the one to keep in your head all day. Opinions are worthless. What someone does is data.

So on Friday, you're not running a focus group and you're not collecting compliments. You're watching five people use a thing and writing down what actually happens to them.

The whole team watches, against one grid

Here's the move that turns five conversations into a decision instead of five different memories of five conversations.

The whole team watches every interview. Live. Usually piped to a second room so the customer only sees the one interviewer, but everyone else is in the building, eyes on the same screen. And they're not just watching. They're taking notes against a shared grid.

Build the grid before the first customer walks in. Down one side, the questions you came to answer, the bets from Monday you're trying to confirm or kill, written plain. Something like "can they find the export button?" Across the top, your five customers. Every watcher fills the same boxes as they watch. What happened, where it broke, what surprised them. So the box for that question under customer three reads "hunted for forty seconds, gave up, asked where it was." Same structure, same prompts, so by the fifth interview you're not comparing your gut to someone else's gut. You're comparing the same boxes, filled five times.

Why bother? Because memory lies and it lies in your favor. Left to recall it, every person on the team remembers the moments that confirm what they already believed, and the room walks out "agreeing" on five different things. The grid is the cure. It forces everyone to write down what they saw while they saw it, in the same shape, so the read at the end belongs to the evidence and not to whoever talks first. It's the same instinct as the silent voting in a Decide sequence. Structure the input so the loudest person in the room doesn't quietly become the answer.

Look for what repeats

After the fifth interview, you don't argue. You look at the wall of grids and you hunt for patterns.

One person getting stuck on one screen is noise. It might be them, it might be a bad night's sleep, it might be nothing. The same screen, the same wall, three or four of the five customers? That's not noise. That's a finding. The thing you're looking for is repetition, the problem that shows up over and over across different people, because repetition is what tells you the issue is in the prototype and not in the person.

This is exactly Nielsen's 85% showing up in real time. By the third or fourth customer, you start seeing the same walls, the same confusions, the same moments of delight, and the fifth mostly confirms what four already told you. When that happens, the test worked. You found the real problems, and you found them on five people instead of fifty.

Cluster what repeats. Three or four clear patterns is a good day. Those patterns are your answer. This part of the bet held, this part broke, here's the next thing to fix or rethink.

When to reach for it

Friday is the back half of a Design Sprint, the bridge from "we built a prototype" to "we know whether it works." But the five-customer test isn't trapped inside the five-day week. Pull it out and run it any time you've got something real enough to put in front of a person, a clickable prototype, a fake front, even a rough flow, and you want an honest read before you commit real money to building it.

That makes it a lean experiment in its own right, the cheap test you run to raise your confidence before the expensive build. You need a prototype to test, a handful of real customers who match who you're building for, a quiet hour each, and a team willing to watch instead of guess. That's the whole kit.

Reach for it whenever the question is "can a real person actually use this, and do they want to?" and you'd rather find out from five people this week than from the whole market after launch.

What good looks like

You'll know it worked by how the room feels at the end of the day, and by what it does next.

Nobody had to argue about what they saw, because everybody wrote it down in the same boxes while it was happening. You walk out with three or four clear patterns instead of five contradictory opinions, and a team that agrees on what's broken because they all watched it break.

And the answer is honest, because it came from behavior. Not from a customer being nice to you. From five real people trying to use a real-feeling thing, and you reading their hands instead of their compliments.

That's the trade Friday buys you. One day, five people, and a straight answer about your idea, before you pay to build it for everyone.

Credit where it's due: the design sprint, the Friday test day, and the five-act interview structure are Jake Knapp's, with John Zeratsky and Braden Kowitz, from Sprint (GV / Google Ventures, 2016); Knapp credits the five-act interview to Michael Margolis, GV's research partner. The "test with about five users to find roughly 85% of usability problems" finding is Jakob Nielsen's, Nielsen Norman Group, in his 2000 essay "Why You Only Need to Test with 5 Users," built on the mathematical model he and Tom Landauer published in 1993. The behavior-over-opinions discipline is the heart of The Mom Test by Rob Fitzpatrick. What's mine is how I run the room and when I reach for this test outside the full sprint.