The 30-Day AI Champion Bootcamp · Lesson 3 of 4

Week 3 — Test with One Real User

> 30 min/day · 6 days · A 5-minute Loom showing before/after by Saturday

You built a thing. Now one real person who is *not you* is going to use it while you sit quietly and watch.

This is the uncomfortable week. It's also the week that separates the champions who get funded from the ones who don't. Leadership has seen thousands of demos. They have not seen many real users honestly interacting with prototypes. If you walk into the pitch with that recording, you're playing a different game.

The one user rule

You're testing with one coworker — ideally the person you interviewed in Week 1. Not "a few users." Not "a small pilot group." One.

Why one? Because:

  • You can find bugs with one user. You don't need a sample size.
  • Leadership trusts a named coworker more than an anonymous "8 respondents."
  • Fixing a handful of real things for one real person is a week of work. Fixing them for five is a month.

Pick the most generous person from your Week 1 interviews. The one who said "oh yeah, this takes forever, I'd love anything that helps." They're rooting for you. Use that.

Your week, day by day

Day 1 — Ask (30 min)

Slack or email your user. Short:

> *"Hey — I built the thing we talked about. It's rough but it works on the [invoice/timesheet/warranty] process we discussed. Could I grab 30 minutes with you this week to watch you try it? I want to see where it breaks. I'll bring coffee."*

That last line is half joke, half important. Make it easy to say yes. Offer a calendar slot. If they wave you off with "just send me a link and I'll try it," politely insist on being there live. Watching is the point.

If this person says no — go to your second-choice interviewee from Week 1. Don't skip the live-watch.

Day 2 — Prepare the session (30 min)

Open the User Test Observation Sheet template. It has columns for:

  • What they tried to do
  • What they actually did (word for word if you can)
  • What went wrong
  • What they said out loud ("huh," "wait," "why did it do that?")
  • Your notes / what to fix

Have two tabs ready: your prototype, and the "normal" way they do it today. You're going to watch both.

Do NOT prepare a slide. Do NOT script an introduction. You're not presenting — you're observing.

Day 3 — Run the session (30 min)

This is the hardest 30 minutes of the bootcamp.

First 5 minutes: Let them do the task the normal way, start to finish. Time it. Write down everything they open, type, copy, paste. This is your "before."

Next 15 minutes: Hand them your prototype. Say:

> *"I'd love it if you just tried to do the same task using this tool. I'm not going to explain it — I want to see what's obvious and what's confusing. Talk out loud while you're doing it if you can."*

Then do not speak. Do not hint. Do not help them find the button. If they ask "wait, what am I supposed to do here?" you say "whatever feels right — I want to see." And write that down.

Last 10 minutes: Debrief. Ask:

  • What felt better?
  • What felt worse than just doing it yourself?
  • If you had to use this every day, what would drive you crazy?
  • What one thing would make you actually use it tomorrow?

Thank them. Genuinely. Buy the coffee.

Day 4 — Process what you heard (30 min)

Open your notes. You're going to have a list of maybe 8–12 things that went wrong or felt bad. Sort them into three buckets:

  • Must fix before anyone else sees this — the things that made them visibly confused or angry
  • Should fix for v2 — minor annoyances worth noting
  • Not going to fix (on purpose) — scope creep, out of bounds for this demo

Pick two from bucket 1. Only two. That's your Day 5 work.

Day 5 — Fix the two things (30 min)

Go back to your prototype and fix the top two issues from the test. Don't redesign. Don't add features. Surgical fixes only.

Examples of "two-things" fixes:

  • Add a starter example so the user sees what to type
  • Make the bot explain what it just did (not just output the answer)
  • Tighten the system prompt to never output the one thing your user hated
  • Add a "try again" button
  • Rewrite the one confusing label

Test each fix with the same real input you used before. Make sure the fix stuck.

Day 6 — Record the Loom (30 min)

Now the moment this whole week was building toward.

Open [Loom](https://www.loom.com) (free tier is fine). Record a 5-minute video in three parts:

Part 1 (60 seconds) — Context "Hi, I'm [name], I work in [dept]. I spent the last month building a prototype for our [problem]. Here's what's happening today [show the manual process, sped up]. Today, [user's name] spends about [N] minutes on each one, and they do [M] per week. That's [hours/year]."

Part 2 (2.5 minutes) — The Before and After "Here's the tool I built. Watch what happens when I feed it a real [invoice / email / document]." Demonstrate one real input. Show the output. Show the time comparison. If you can splice in 20 seconds of your Week 1 test where the user reacted to it working — even better.

Part 3 (90 seconds) — The Ask "The prototype is working. [User] tried it last week. What I'd love to do next is [the scaled-up version — roll it out to the 3 people doing this, or automate the email step too]. For that I need [specific ask — 4 hours of IT time / a license / a one-hour meeting with [exec]]."

That's it. Five minutes. No slides yet — that's Week 4.

Rough is fine. It should NOT look like a marketing video. If you re-record it more than twice, you're over-polishing.

What "done" looks like
  • You watched one real person try your prototype
  • You fixed the two biggest issues they found
  • You have a 5-minute Loom with before / after / ask
  • You know exactly how much time this saves per use, per person, per week
  • You have a quote or a moment you could cite from the user in the pitch
The templates this week
  • User Test Observation Sheet — the silent-watch note-taking grid
  • Fix Triage Matrix — the three-bucket sort for everything they found
  • 5-Minute Loom Script Outline — the three-part script structure
What you've just learned

You just did what 90% of "AI champions" never do: watched a real person use their thing and had the discipline to not help them. That recording, that one user, that named person in your company — this is what separates proof from PowerPoint.

Move to Week 4 — time to turn all of this into a 5-slide deck and a meeting on the calendar.

Previous: Week 2 — Build the Prototype · Next: Week 4 — The Pitch