Pro edition

Research Plan.

Sent in week zero, before anything is scheduled. Most research fails after it is done rather than while it is being done, so half of this document is about what happens to a finding between the session and the decision.

How the sessions run

Five participants, one hour eachFive is not a sample and it is not pretending to be. It is the number at which the same problem starts appearing twice, and the second appearance is what makes it a finding rather than an anecdote.
Recruited from your users, not a panelA panel gives you people who are good at being research participants. Your support inbox gives you people who are annoyed about something specific, and they are worth five panellists each.
Recorded, transcribed, quotedQuoted rather than summarised. A summary is where a finding goes to become whatever the reader already believed, and it happens without anybody intending it.
Two observers maximumOne from the team, one from whoever will build it. Three or more and the participant is performing for a room.
Ten minutes of nothing at the startWhat did you do this morning. No product, no screens. Everything useful in these sessions has come out after the participant stopped answering carefully.

How findings get written

A finding has to survive three weeks, four retellings and one person who disagrees with it. Almost none of that is about the research and all of it is about the writing.

One page, not a deckWhat I think is true, what I am unsure of, what I would need to find out. A deck is a performance and it gets skimmed; a page gets read and argued with, which is the objective.
Confidence stated per findingObserved in 4 of 5 sessions. Observed once, and it was striking. Inferred, not observed. Three different things and they get written differently.
The quotes stay inVerbatim, with the ums. A cleaned-up quote is a paraphrase wearing quotation marks.
What would change my mindWritten down before anybody disagrees, so the disagreement has somewhere to go that is not a meeting.
Circulated before the meetingTwo days before. A finding read in advance gets challenged on its merits; a finding presented live gets challenged on whether the presenter seems confident.

Six ways this goes wrong

I have done the third one on a project that is on the front page of this site, and it cost three weeks.

The summary drift: Nineteen sessions become a five-bullet summary become one sentence in a deck become a feature. Nobody lied at any step. The fix is that the quotes travel with the finding all the way to the end.
The loudest participant: One articulate person who described the problem well becomes the whole research. They are usually right about their own problem and wrong about how common it is.
Research as confirmation: Running the sessions after the design is drawn, which turns a study into a review. I did exactly this on the clinic project and it cost three weeks.
The stakeholder in the room: A participant will not criticise a product in front of somebody who obviously built it, and will do it freely twenty minutes later to a stranger.
Counting the wrong thing: Time-on-task falls and everybody celebrates, and the reason it fell is that people gave up. Every efficiency metric needs a completion metric beside it or it is not a metric.
Findings with no owner: A finding nobody is responsible for acting on is a finding that gets rediscovered in eighteen months by whoever comes next, at full price.

What you keep

All of it, including the versions that did not work
Artefact	Whose	Note
Session transcripts	Yours, with participant consent	Raw. Redacted only where a participant asked.
The one-page findings	Yours	The document the project is actually steered by.
Annotated flows and screens	Yours	Including empty, error and loading states, which is where half of the real design work is.
The discarded versions	Yours	In the same file, not deleted. Whoever picks this up in two years needs to know what was already tried.
A two-hour handover	Booked before the last week	With whoever builds it. Worth more than every annotation combined.