How I research (notes from the field)
Research isn’t a phase that ends when the deck ships. It’s weekly contact with the people who feel the problem — engineers staring at a failed run, agents on a support call, patients hunting for a clinic, testers walking a prototype the same night. The product discovery process I trust starts with behavior change as the outcome, not a feature list. Continuous discovery means discovery and delivery run in the same week: notes come in, clusters form, the next bet gets sharper, and something ships that can teach us.
Where the notes come from
I match the channel to the risk. Engineer sessions and support threads feed Artificial Chip Intelligence work at Cognichip — the first designer there, so I hear the failure modes before they harden into tickets. At Plume, support tickets and NOC escalation notes told the truth about what agents faced when RF interference, bad pod placement, and a weak client radio all looked like the same angry call. Boxbee interview sessions landed in Dovetail, tagged as they came in. UI/UF same-night tester notes hit a shared Miro board before the room emptied. For institutional IA at UCSF, tree tests and card sorts showed how people actually look for care — by condition and location, not by the org chart.
The method isn’t sacred. The question is: what evidence would change what we build next?
Clustering without losing the quote
Raw notes are noise until they’re stories with receipts. I use synthesis tools to cluster interview notes, tickets, and transcripts into opportunity themes — with the quotes still attached. A theme without a quote is a hunch wearing a lab coat. Human judgment decides which stories are real versus loud, and who to talk to next. Loud complaints often point at symptoms; quiet, repeated friction usually points at the opportunity.
That research synthesis workflow is fast on purpose. Same-day clusters at UI/UF meant each designer left with a findings sheet instead of a stack of stickies. At Cognichip, session notes become opportunity cards before the week turns over. Speed only helps if the quote survives the roll-up.
Quotes
- “I rebooted twice already”
- “I can’t find which box has my ski boots”
- “I want to lend a box to a friend”
Clusters
- RF interference vs bad placement vs a weak radio
- Find the box
- Lend the box
Opportunities
- First-call resolveRiskiest assumption: the agent trusts a ranked cause without another reboot.
- Two asks, two opportunitiesDifferent opportunities, even in the same interview.
From stories to opportunities
Opportunity mapping follows a simple discipline: an opportunity should have more than one possible answer. If there’s only one solution hiding in the wording, it’s a solution in disguise — rewrite it until alternatives exist. Next to each opportunity I write the riskiest assumption: desirable, viable, feasible, usable, ethical. That assumption is what the next prototype has to answer, not the whole feature.
Customer interview synthesis feeds an opportunity–solution tree, not a backlog of requests. “I can’t find which box has my ski boots” and “I want to lend a box to a friend” are different opportunities even when they show up in the same interview. Separating them keeps discovery honest and keeps delivery from shipping the loudest ask.
Methods matched to risk
Design research earns its keep when the method fits the decision:
- Tree test / card sort when institutional IA is the risk (UCSF): find where people get lost before anyone styles a pixel.
- Incident replay when ops tools are the risk (Plume dashboards): a triage queue that only works on a clean happy path fails in the first week of an outage.
- Same-night severity clustering when community research is the risk (UI/UF): realistic tasks, new eyes every round, synthesis before the room empties.
- Tagged interview sessions when lifecycle and state are the risk (Boxbee): patterns roll up into an opportunity map that points the next design round.
Match research method to risk. Prototypes answer a specific question. Empty, error, edge, and accessibility states count as research when they’re the places people actually get stuck.
Handoff into framing
Clusters don’t sit in a folder. They become the input to outcomes, bets, and decision rules. An opportunity map with quotes attached is what I hand into framing: what behavior should change, what we’re betting, and what evidence would make us stop, revise, or scale. That handoff is the bridge from continuous discovery into product thinking — the Sense → Frame sequence without the stage theatre. Discovery and delivery dual track means Monday’s bet can already smell like last week’s calls.
What stays human
Tools cluster. People decide. Who to interview next, which theme is noise, which assumption is actually the riskiest, and whether something clears ethical, usable, and viable filters — those calls stay with me and the trio (design, product, engineering). Partnering with Product and Engineering on what a threshold change means before it reaches a customer is part of the research job, not a later politics step. Clarity out of ambiguity starts here: one clear next question, then one clear next action.