user-research-synthesis

User Research Synthesis

Two failure modes dominate bad research synthesis: forcing raw data into categories decided in advance, and asking interviewees what they want instead of reconstructing what they actually did. This skill covers two named, well-established techniques that avoid both — affinity mapping (the KJ method) for synthesizing a batch of observations into themes, and Jobs-to-be-Done "Switch" interviewing for understanding a specific past decision.

Affinity mapping (the KJ method)

Developed by Japanese anthropologist Jiro Kawakita, and adopted widely in UX research as a bottom-up alternative to top-down categorization. The entire point is that themes emerge from the data instead of being decided before you've looked at it.

Procedure

1. Atomize first. Before mapping, break every source (interview transcript, support ticket, survey verbatim) into individual, single-idea observations — one sticky note (physical or virtual) per idea. A transcript quote covering three separate points becomes three notes, not one. Mapping breaks down immediately if a note carries more than one idea, because it can't be sorted into a single cluster.
2. Silent individual clustering first. Each participant in the session moves notes into groups without talking. This is not a formality — discussing groupings while forming them causes the most vocal or senior person's mental model to become everyone's mental model before the data has been seen fresh. Silence is what makes the bottom-up property real.
3. Merge and negotiate second. Once individuals have independently clustered, compare groupings as a group, discuss disagreements, and merge into a shared set of clusters.
4. Name clusters last, after they exist. Write the theme label only once a cluster has taken shape from the notes inside it. Naming a cluster before sorting notes into it (or pre-writing category headers before the session starts) is functionally indistinguishable from deductive coding against a preset framework — it defeats the reason to use this method over a spreadsheet with columns.
5. Look for outlier notes. A note that resists joining any cluster is informative, not a defect — it often flags a segment or edge case the rest of the data doesn't represent. Don't force it into the nearest-fitting group just to close the session.
6. Write the synthesis from the clusters, not from memory. The output is a named theme plus the observations that support it, so a reader can trace every claim back to raw data — a synthesis with no traceable observations underneath it is an opinion, not a finding.