3.3 Contextual Inquiry

In Brief

Description

In contextual inquiry you visit the participant’s actual environment, watch the work happen, and ask questions in the moment rather than after the fact. You model an expert/apprentice relationship: the participant is the expert in their own work, and you are the apprentice learning it.

Observation produces different data than interviews because participants can only describe what they consciously do; they cannot describe the workarounds, sticky-note hacks, and tacit shortcuts they don’t notice themselves doing. A retrospective interview gets you the participant’s mental model of their work. A contextual inquiry gets you the work itself — including the parts the participant would never think to mention. Workarounds, environmental constraints, and interruptions tend to appear only when you watch the work happen.

Contextual inquiry rests on four principles that describe how you behave on-site. Context means going to the real place where the work happens, not a conference room. Partnership means treating the session as a shared inquiry — the participant drives the work, you follow and ask. Interpretation means voicing your reading of what you just saw and inviting the participant to confirm or correct it. Focus means having a defined research question so you notice what matters and don’t drown in everything.

The output is qualitative and small-sample. It produces hypotheses, not statistics — pair it with a downstream evaluative method (Paper Prototyping, Comprehension Test, closed-ended survey) once the field data has shaped a hypothesis worth testing.

How to

Prep AI Prompt

1. Define your research question. Create one clear research question. For example, “How do mid-market accountants actually close the books at month-end?” “Learn about accountants” is not sufficiently focused. The question shapes what you’ll watch for and which workflow steps you’ll ask the participant to walk you through.

2. Recruit five or more participants in their real environment. A contextual inquiry done at the participant’s desk during their actual workday produces different data than one done over coffee. If the problem only happens at certain times (month-end close, surge periods, mornings), schedule the visit for then. Recruit at least five participants — small samples mean any single observation is anecdote until a pattern repeats.

3. Train the observer pair. Pair the lead researcher with a second observer when possible — one focuses on the work, the other captures notes and timestamps. Both should agree in advance on what counts as a workaround worth probing. If you’re solo, plan to record the session so you can re-watch instead of trying to take notes mid-observation.

4. Prepare a framing statement. Two or three sentences you read at the start of the session: who you are, what you’re trying to learn, that you’ll mostly watch, and that you may interrupt occasionally to ask what they just did. Make clear they’re the expert and you’re the apprentice — something like “I’d like to watch you do [task] the way you normally would. I’ll mostly watch and ask a few questions along the way.”

5. Prepare an observation checklist, not a script. Unlike an interview, you’re not running a list of questions. You’re going to watch real work. The checklist is a reminder of which workflow steps and decision points you want to make sure get covered if they don’t surface naturally. Include space for: workarounds noticed, tools or substitute products in use, interruptions, body language during difficult moments.

6. Get permission to record. Audio at minimum, video or screen-share if it doesn’t change behavior. Promise the participant you’ll send them a summary and ask before quoting them anywhere.

Analysis AI Prompt

1. Debrief while it’s fresh. Within an hour of the session, sit down with your notes and the recording and write up: the workflow as observed, every workaround noticed, every place the participant hesitated, every place the environment intruded. Two hours of debrief per session is normal.

2. Run an interpretation session. With your team, walk through each session’s notes and surface every observation as an explicit interpretation: “When the participant copy-pasted the report into a spreadsheet, that means the built-in sort is unusable for their workflow.” This is the team’s chance to challenge each interpretation. Disagreement means you don’t yet know what you saw and need to look at the recording or follow up with the participant.

3. Group similar observations together. Take every observation, every workaround, every interpretation across all sessions and put each on its own sticky note (physical or digital), then group them by what they have in common, not by which session they came from — a technique called affinity diagramming. The clusters that emerge are the patterns. Single-session observations that fit no cluster are anecdotes — note them, but don’t build on them.

4. Identify workflow patterns. Look across the clusters for repeated structures: the same workaround appearing in different sessions, the same decision point repeatedly going wrong, the same environmental constraint appearing across different participants. A pattern that shows up in three of five sessions is a pattern. A behavior that shows up in one session is a hypothesis worth probing further.

5. Be honest about sample size. Five participants is enough to surface patterns; it is not enough to claim those patterns generalize. State what you observed, in how many sessions, with what variation. Resist extrapolating qualitative observations to the entire population — write findings as “in five sessions we saw X” rather than “users do X.”

6. Synthesize into a hypothesis worth testing. The output of analysis is one or two clear hypotheses for downstream evaluative testing — a paper prototype to test against a workaround you discovered, a closed-ended survey to size how widespread the workflow pattern is, a comprehension test on a value proposition you can now articulate. Contextual inquiry doesn’t end the research; it sets up the next round.