Unit 2: Discovery for AI features: techniques that work when users can't imagine the solution

Learning objectives

Adapt continuous discovery techniques for AI product contexts. Use observation and prototype testing when users can't articulate needs. Identify AI opportunities from behavioral data and workflow analysis.

Video script

Reading material

Combining techniques for AI discovery

The strongest AI discovery insights come from triangulation: combining multiple techniques to validate the same opportunity.

Start with behavior analytics to identify candidate opportunities. Then do workflow observations to understand the context around those behaviors. Then build a prototype and test it to see whether users would actually adopt the capability. Then use interviews to understand the nuance and refine the approach.

Each technique fills gaps in the others. Analytics tells you what's happening but not why. Observation tells you why but is slow and small-scale. Prototypes tell you if users would adopt the solution but not if you've identified the right problem. Interviews fill in the reasoning but are subject to articulation bias. Together, they give you a much clearer picture.

The "jobs-to-be-done" lens for AI features

The JTBD framework is particularly useful for AI discovery because it focuses on what users are trying to accomplish rather than what features they want.

When a user "hires" your product to do a job, AI can potentially do that job faster, better, or more automatically. The discovery question becomes: which jobs are users struggling with, and would AI assistance make the job noticeably better?

The key word is "noticeably." If AI saves users 30 seconds on a task that takes 5 minutes, they'll probably notice and value it. If AI saves them 5 seconds on a task that takes 30 seconds, they probably won't, and the overhead of interacting with the AI feature (reviewing the output, correcting errors) might negate the savings entirely.

Practical exercise

Exercise: AI opportunity identification

Choose a product you use daily for work. Conduct a self-observation exercise.

1. Track your own workflow for one hour. Note every repetitive action, every context switch, every moment of friction.
2. From your observations, identify 3 tasks that an AI could potentially handle or assist with.
3. For each task, answer: how much time would AI save? Could AI do this well enough that you'd trust it? What would go wrong if the AI made a mistake?
4. Rank the three opportunities by a simple impact/feasibility assessment.
5. For your top opportunity, describe what a prototype would look like. What's the simplest version you could test?

Write up your findings in a one-page AI opportunity brief.