Foundational Research — 2025
Finding where AI belongs — and where it doesn't
The instruction came down as one word: agentify. No problem attached, no user in the sentence. Pioneer research turned an open directive into a prioritized, evidence-backed portfolio of use cases — and, just as usefully, a map of the places users wanted left alone.
Research lead — foundational research, opportunity mapping, use case prioritization
- use cases identified
- 11use cases identified
- already moving into build
- 4already moving into build
The problem
Senior leadership instructed the team to agentify a digital asset. What that should mean for the people using it was an open question.
The asset had no AI anywhere in the experience, so there was no baseline to improve on and no evidence about where an agent would be welcome.
- 01
No AI, no personalization
The experience offered nothing adaptive. Every user met the same static interface regardless of context or history.
- 02
No AI-assisted retrieval
Users could not ask an AI for information and get an answer. Finding anything meant knowing where it lived already.
- 03
No AI navigation
Nothing guided users to the right place. Wayfinding was entirely on the person, every time.
The approach
Before deciding what to build, I set out to find where AI would actually help. This was pioneer research on an untested question: not “how should the agent work,” but “where in this asset does an agent earn its place?”
Where is the user's effort disproportionate to the value they get back?
Four lenses
- 01
Pain points
Where the current experience breaks down, frustrates, or stalls the user outright.
- 02
Time-consuming tasks
The work that costs users the most time — the clearest candidates for compression.
- 03
Workarounds
What users built for themselves outside the experience. Every workaround marks an unmet need.
- 04
AI-suited tasks
Where users said an agent would genuinely make the task easier, in their own terms.
What I discovered
The finding that mattered most was where AI does not belong. Mapping sensitivity turned out to be as valuable as mapping opportunity — an agent placed where users don't want one costs trust across the whole digital asset, not just that task.
High receptivity
Let the agent carry it
- Finding information without knowing where it lives
- Repetitive, high-effort tasks users already dread
- Tasks users themselves named as easier with AI
Sensitive territory
Keep the human in front
- Actions that involved the handling of sensitive data were explicitly flagged by participants as risky candidates for an AI solution
When sensitive data is involved, users wanted control.
“The hardest part of an AI mandate is deciding where not to use it.”
The outcome
Eleven discrete use cases, prioritized. Four moved into build immediately, with the rest queued behind them and the sensitive territory fenced off before scoping began.
A mandate became a roadmap
An open instruction to agentify turned into a ranked, evidence-backed portfolio leadership could act on.
Sequencing grounded in evidence
Which three go first was answered by user effort and receptivity, not by what was easiest to ship.
Guardrails set early
Sensitive actions were flagged as risky before a single agent was scoped into them.
A reusable frame
The same four lenses can assess the next surface without starting the research over.
The takeaway
Opportunity and sensitivity: mapping where users resist AI protects adoption as much as mapping where they welcome it.
Findings leadership can act on: eleven discrete, prioritized use cases — a format that converts directly into roadmap decisions.
