Signing in

You will be sent to MillerKnoll sign-in.

AI

Running Team Experiments That Actually Work

Most AI experiments fail because the experiment was not designed to produce learning, activity without insight.

Lesson 3

Friction first, not features.

Start with genuine friction (slow, heavy work), not "find something for AI." Test small: one person, one task, two weeks.

Team experiment flywheel

Notice frictionTest smallDebriefDecide scale
Notice friction → test small → debrief → decide what scales.

Core principles

  1. Good candidates: repetitive, documented output, bounded scope, low-stakes, currently suboptimal.
  2. Poor candidates: highly variable tasks, high-stakes failure cost, outputs you cannot compare.
  3. Debrief (20 min): what we tried, what we observed, what we learned (specific), what we do next.
  4. Example learning: "Recap saved 20 minutes but two action items were wrong, we verify names before sharing."

Check yourself

What makes most AI team experiments fail, according to this lesson?

Do this in Copilot

Pick one team friction; design a two-week experiment (who, task, AI approach, how you evaluate). Do not launch until Lesson 4 if champions will support it.

Paste this into Copilot Chat and work through it before moving on.

Experiment design

We want to test Copilot for [TASK] for two weeks with [ROLE]. Draft: hypothesis, success criteria, what we will observe, and debrief questions, one page max.
Open Copilot →
  • Experiment design

Did you run this in Copilot? Mark complete when you have tried it.

Next lesson: Finding, Developing, and Not Burning Out Your AI Champions →