Four designers on Apple TV’s enterprise tools were running their own research, each in their own way. I gave them a shared framework and common tools, built on the AI skills my team had already governed. The first study to use them was a concept test for a new rights management platform. Six findings went into the next design round, and the designer who moderated kept the skills.
01 — Context The designers hadn’t waited for a researcher
Four designers each covered a different product and ran their own research, with limited shared framework or common tools.
The design was already moving. The design lead had built a clickable prototype with AI, informed by an earlier contextual inquiry, and the product team wanted their picture of the business process checked against real users. Working with a researcher was new to everyone involved, so there was some skepticism to begin with. Research had to run alongside the design, and there was roughly one study’s worth of time across four designers and four products.
The choice
I could run one good study for them, or leave them with a way to keep running their own. Only the second one survives me moving to the next thing, so the study had to double as training.
I wrote the plan, the session guide and the findings report. A second designer moderated (deliberately not the design lead, who had built the prototype), and I coached her through facilitation because she wanted the tools for her own research afterwards.
02 — Study Did our picture of the process match the work?
That question made the documented business process the subject of the study, and the prototype the way into it.
The planning and synthesis skills drafted and clustered. I decided what the findings were. Before any participant saw the prototype, I ran synthetic users against it to see what a walkthrough would catch on its own.
Only one of the 22 was a defect worth fixing first. The rest were predictions the sessions could check.
Why synthetic first
A stale record or a mislabeled button is findable without a participant, and finding one in a session costs an hour I can’t get back.
How the synthetic pass worked
I gave Claude the background first: what the platform is for, who uses it, and what the team was trying to decide. The synthetic users were built from the personas and journeys the design lead had already mapped. For the prototype, Claude got the HTML plus screenshots of every modal and dropdown, since it can’t click through it. Then it ran one walkthrough each, as a sales executive and a legal counsel, reacting to the interface in character.
Then four real sessions
Moderated concept testing with a clickable prototype, four participants recruited through the product owner. Four fit the sprint: enough to validate the concept and surface the patterns worth acting on, with a further round to follow. Each session opened on the documented business process so participants could mark where their own work diverged from it, then moved to the prototype. The product team sat in as observers.
03 — Findings What four sessions turned up
There were six. The last is the one the product team had asked us to check, and it wasn’t the answer they expected.
What the walkthrough couldn’t see
The synthetic run had produced 22 findings before anyone sat down with a participant, eight from the sales persona and fourteen from the legal one. Set against the findings report:
The two personas it ran as were the roles the design already assumed, so the broadest of the six was never available to it. Synthetic testing is only as good as the persona you can describe. As a first pass it earned its place, as long as someone with domain knowledge reads the output and catches what it invents.
04 — Outcome The process gaps went to change management
Ten recommendations went to the team, split into strategic, near term and quick wins, each tied to the findings behind it. The business process gaps went over as change management work, which the team hadn’t expected to be handed.
After the study
05 — Evidence The documents
The plan and guide, the findings report, and both synthetic walkthroughs, in full. They’re redacted: product and workflow names are generalized, colleagues appear as their function, and the interface screenshots are blurred.
06 — Reflection A small study was a good place to pilot the skills
The team was already adopting AI, so bringing our research skills in made sense.
If I did it again, I’d work more closely with the product managers from the start, to get them on board with the designers taking on the research practice. I think that would have eased them into research and helped them buy into the process and the methods sooner.