Design thinking for AI projects
Design thinking gives a group a way to understand a problem before choosing a solution. For AI projects the steps are familiar. What changes is what you look at while you run them.
What changes when the solution is AI
A workshop about an AI project can use the same methods as any design thinking workshop. The difference is in six things the group should look at and say out loud.
| Look at | What to ask in the workshop |
|---|---|
| Who the work is for | The people who use the output and the people who answer for it, not only the user of the tool. |
| What the problem is | The work and the outcome. Avoid naming AI in the problem, because AI is one possible answer. |
| What could go wrong | The cost of an error is part of the design, not a test at the end. |
| What the data is | Whether the data to do the work exists, is accessible and is trusted. |
| What stays human | Who checks, who decides and who answers. Decide it early. |
| The alternative | Whether a simpler, non-AI fix would do. Always write it down. |
The six phases of a workshop
You rarely run all six phases at once. Start where the group is stuck, and expect loops. A test often sends you back to Frame.
- Understand: learn how the work really happens. You leave with a picture of the people, the work and the pain.
- Frame: agree the problem. You leave with a problem statement and questions to answer.
- Ideate: generate options. You leave with many ideas, then a few idea cards.
- Select: choose what to take forward. You leave with ideas placed and ranked on agreed criteria.
- Prototype: make it tangible. You leave with something people can react to.
- Test: learn before building. You leave with what held, what did not, and what changes.
Eight questions to ask in every session
Keep these on the wall and ask at least three in every session, whatever method you are running.
- Who answers for what the AI solution produces?
- What happens if it is wrong, and how bad is that?
- What data does this need? Does it exist, and can we use it?
- Which part must stay human, and why?
- What would a simpler, non-AI fix look like?
- Who is affected but not in the room?
- What would we need to see to trust it?
- How will we know it worked?
Which method for which situation
| If this is the situation | Start with | Minutes |
|---|---|---|
| Nobody agrees who is affected | Stakeholder and Impact Map | 40 |
| Assumptions about how people feel | Empathy Map or Interview Guide | 30 to 45 |
| The real process is unclear | Current-State Workflow Map | 60 |
| The problem keeps being described differently | Is / Is-Not, then Problem Statement | 30 + 30 |
| The problem keeps coming back | Five Whys or Fishbone | 20 to 40 |
| The group will not speak freely | Write, Pair, Share | 20 |
| You need many ideas fast | Structured Brainstorm or Crazy 8s | 15 to 30 |
| An idea is vague | Idea Card | 25 |
| You need to narrow many options | Value and Feasibility Placement or Dot Voting | 15 to 30 |
| You need to test before building | Pre-mortem or Experiment Card | 25 to 30 |
For a recommendation that fits your group size, time and setting, use the method finder.
Where to start
The free Workshop Toolkit has a facilitator guide and a printable canvas for each method in the first four phases. To run a session, read how to run a design thinking workshop.
Questions
What is design thinking for AI projects?
It is the same human-centred approach to problem solving, applied to projects where the solution may involve AI. The methods are familiar. What changes is what the group looks at: who answers for the output, what the cost of an error is, whether the data exists, what stays human, and whether a simpler non-AI fix would do.
Do I have to run all six phases?
No. You rarely run all six at once. Start where the group is stuck. A test often sends you back to Frame, so loops are normal.
Should the problem statement mention AI?
Avoid it. Describe the work and the outcome. AI is one possible answer, and naming it too early closes off simpler alternatives.
When should a workshop write down a non-AI alternative?
Every time. Ask what a simpler, non-AI fix would look like and record it, so the group can see whether AI is justified.