Context
Egnyte is an AI-powered content cloud: file storage, governance, and a growing set of AI features that work on a company's own documents. The features worked. Inside the company itself, though, plenty of people never touched them, and nobody could say why. Anecdotes pointed everywhere at once. The features were hidden. The outputs fell short. People distrusted AI. People had no time.
Problem
Low usage is a number. It doesn't tell you which problem you have. Awareness, access, output quality, and trust all produce that same number, and each one calls for a completely different product response. My job was to isolate the actual cause, rigorously enough that the product team could act on it.
What I did
I designed the study end to end: recruiting, interview guides, survey instruments, analysis, synthesis. Twenty-three employees across eight departments and three seniority tiers, so the sample covered the full range of how work happens at the company.
The decision that made the study work was segmentation. I grouped participants by observed adoption behavior: regular users, occasional users, and non-users. Then I looked at what actually separated the groups. A role-based cut would have shown which teams use AI. The behavioral cut showed why some people adopt and others don't, with job function held aside.
Twenty-three employees, eight departments, three seniority tiers, grouped by what they actually did with the AI rather than where they sat.
- 23Participants
- 8Departments
- 3Seniority tiers
The analysis was thematic coding of interview transcripts and open survey responses, with each theme tracked against the behavioral segments. I scored every theme on three axes: how often it appeared, how hard it blocked adoption when it did, and how directly the product could address it. Every theme ended up with three numbers beside it and a place in the ranking.
Method · behavioral segments
How the three adoption segments were defined, and what separated them. Opens full size.
What I found
The primary driver was discoverability. People who ran into the AI features during their normal work used them and kept using them. Most people never ran into them at all. The gap between users and non-users was mostly a gap in exposure. Exposure is cheap to change. Rebuilding the underlying AI is not, and the data did not call for it.
Finding
Adoption was a discoverability problem, not a capability problem. The people who found the features used them.
What changed
I synthesized the findings into a prioritized five-agent roadmap, with each item grounded in a specific research theme and ranked by the scoring above, and presented it to the product organization. It informed AI product strategy for the following quarter.
Three choices in the study design did most of the work. Segmenting by behavior instead of role. Sampling across the whole company rather than the teams already closest to the product. Scoring the themes rather than ranking them by feel.
Some findings concern internal details and stay private. Full case study available on request.