How do you grade a customer discovery assignment?

Grade the evidence and the change of mind, not the write-up. Five criteria work well: whether the student stated falsifiable beliefs before interviewing, whether they tested the belief that would kill the idea first, whether the interviews are traceable to real named people, whether disconfirming evidence is reported rather than buried, and whether the conclusion actually moved. Presentation quality should carry almost no weight, because AI now supplies it for free.

How many customer interviews should a student assignment require?

For a single assignment inside a semester course, five to eight interviews is enough to produce a real surprise without consuming the term. For a capstone or an accelerator-style program, fifteen to twenty-five is a more honest number. What matters more than the count is that the student talks to people outside their own network. An assignment of five strangers teaches more than twenty friends.

What should a customer discovery rubric weight most heavily?

Disconfirming evidence. It is the single hardest thing to fake and the single most predictive of whether a student learned anything. A submission where every interview supported the original idea is either a fabricated set, a set of leading questions, or a set of interviews with friends. All three deserve the same grade.

Can you tell if a student used AI to write their customer discovery report?

Not from the prose, and you should stop trying. Detection tools are unreliable and the arms race is unwinnable. Shift to provenance instead: require interview dates, a contactable name for a sample you will actually call, and raw unedited notes alongside the summary. Fabricated interviews are smooth, agreeable, and confirm the thesis. Real ones contradict it, wander off topic, and contain details nobody would invent.

A Customer Discovery Rubric That Survives AI

John Richards · November 2nd, 2026

I taught entrepreneurship at BYU for twelve years and ran the program there. In that time the hardest thing to grade was never the pitch. It was the customer discovery, because what a student hands in and what a student actually did have never been the same document.

That gap used to be manageable. A student who had not done the interviews produced a thin, obviously hollow write-up. You could see it.

AI closed the gap. A student can now generate a complete customer discovery report, with personas, quotes, and a tidy insights section, in about the time it takes to walk to class. It reads better than the work of a student who actually went out and got rejected by eleven strangers. If your rubric rewards the artifact, you are now grading prompt quality.

Here is the rubric I would use instead, and the reasoning behind each line, because you should adapt it rather than paste it.

The principle: grade the change of mind

Customer discovery has one purpose. It is not to validate the idea. It is to find out, cheaply, whether the idea is wrong before the student spends a semester building it.

That means the output you want is not a report. It is a delta: what the student believed on Monday, what they believe now, and the specific thing a stranger said that moved them. A student who ends where they started has either learned nothing or not listened.

So the rubric weights the delta, and treats presentation as nearly worthless.

The five criteria

1. Falsifiable beliefs, stated first (20 points)

Before any interview, the student writes down what they believe about the customer and the problem, in a form that could be proven wrong.

“Small clinics struggle with scheduling” is not falsifiable. “Clinics with fewer than five providers lose more than three hours a week to rescheduling phone calls, and the office manager is the one who feels it” is.

Award full marks only if a reader could design a test that would fail.

Why this is first: it is the only part of the assignment that has to happen before the work, which makes it the one part a student cannot retrofit afterward. Collect it on a separate date, and the rest of the rubric gains teeth.

2. Riskiest assumption tested first (15 points)

Of everything they wrote down, which belief, if false, kills the idea? Did they test that one first?

Students reliably test the comfortable assumption instead. They will confirm that people like the idea before checking whether anyone would pay, switch, or change a workflow. Both are interviews. Only one is useful.

3. Traceable, real interviews (20 points)

Each interview needs a date, a first name, a role, and how the student found that person. For a sample, you need a contactable email.

Say out loud that you will contact some of them. You do not have to contact many. The declaration is what does the work.

Points come off for interviewing friends, classmates, family, and anyone who has a reason to be kind. This is the most common failure mode in undergraduate work and it is not dishonesty. It is that talking to strangers is frightening, and students will optimise around fear unless the rubric prices it.

4. Disconfirming evidence reported (30 points)

The heaviest weight in the rubric, deliberately.

The student must report what contradicted them. Not “some concerns were raised.” Specifically: who disagreed, what they said, and whether it was a real objection or a polite one.

A submission in which every interview supported the original idea gets a low score on this line regardless of how many interviews there were. There are only three explanations for universal agreement, and none of them are good: the questions were leading, the people were friends, or the interviews did not happen.

Tell students this weighting in advance. It inverts the instinct that the assignment is to defend their idea, and that inversion is most of the learning.

5. Conclusion that moved (15 points)

Did the business model change? Did the customer segment narrow? Did they kill it?

Killing the idea, with evidence, is full marks. Say that explicitly in the assignment sheet or no student will believe you.

Presentation, formatting, and writing quality: 0 points. Mention this in class. It is the clearest signal you can send about what the course is actually about.

What a failing submission looks like

Useful to show students, because they rarely recognise it in their own work:

Twelve interviews, all positive, all people the student already knew
Beliefs written after the interviews, in language that conveniently matches what was heard
The pricing question asked as “would you pay for this?” rather than “what do you use now and what does it cost you?”
An insights section that repeats the original idea in new words
No one who said no, and no account of what they said

That submission can look polished. It can be structured, cited, and well written. It still gets a low grade, and the rubric is what lets you defend that to the student and to a department chair.

The provenance requirement, in one paragraph

Do not try to detect AI. You will lose, you will accuse an innocent student eventually, and the time is better spent elsewhere. Require instead: interview dates, raw unedited notes alongside any summary, and contactable names for a sample. Then spot-check two per class, visibly. Fabricated interviews are smooth, agreeable, and on topic. Real ones contain a tangent about the interviewee’s dog, a complaint about something unrelated, and a sentence that stops the student cold. You will learn to see the difference quickly.

Adapting it

For a 15-week course, run this twice. The first round at week four with five interviews, graded gently, and the second at week ten with ten more, graded hard. The first one exists to get the fear out of the way.

For a capstone, raise the interview count and add a line for whether the student returned to someone they had already spoken to. Second conversations are where the real information lives, and almost no undergraduate does this without being told.

For a short workshop or a three-day intensive, drop to three interviews and keep only criteria 1, 4, and 5. The point survives the compression.

Try the ToolSuite free with your class Customer discovery and business model canvas tools built for coursework. Free to start, with academic pricing for programs. Check out our software free →

We built the customer discovery and business model canvas tools in Startup Ignition ToolSuite around this way of grading, because it is how we run our own bootcamp cohorts. Students log interviews with dates and sources, the canvas keeps version history so you can see which assumptions moved, and the scored output gives you something to put in a gradebook that is not a pitch deck.

If you want to talk through how this maps onto your course, including the parts where you think I am wrong, I am happy to. I spent twelve years on your side of this problem.

Back to All News