Micro Case Study—

How I Use AI as a Design Thinking Partner

When working with complex qualitative research, my challenge is rarely a lack of data. It’s maintaining a clear, unbiased understanding of what the research is actually saying while I’m actively designing.

I’ve found that my biggest risk isn’t missing insights — it’s carrying unexamined assumptions forward simply because they feel familiar or operationally convenient.

Problem

As I synthesize interviews, I naturally begin forming mental models about users — what’s different, what matters, what should be distinct.

The issue is that these models often live implicitly in my head:

  • They aren’t always written down

  • They’re influenced by org structure and labels

  • And they’re hard to challenge once design work begins

I wanted a way to surface and test my own framing before committing it to personas or design decisions.

Context

During early synthesis of multi-participant qualitative research, I often reach a point where patterns feel almost clear — but not yet stable enough to become personas or strategy. This is a high-risk moment for premature abstraction.

What I Did

After establishing a Living Research Reference, I used AI to externalize and interrogate my existing assumptions.
Rather than asking AI to generate insights, I used it to retrieve and compare evidence across interviews without resolving contradictions.

I paid close attention to moments of discomfort — where distinctions I expected to hold began to feel thin, repetitive, or overly complex.

Example

While pressure-testing an early persona framing, similarities across participants began collapsing into redundancy rather than clarity. That signal prompted me to pause persona development and revisit my assumptions before committing them to artifacts.

Outcome

This practice allows me to:

  • Commit to fewer assumptions early

  • Be more deliberate about when I abstract

  • Increase confidence that personas reflect evidence, not inertia

AI supports reflection and recall; all interpretation and decisions remain human-led.