Insight

Origin & evolution

The mechanics of turning raw observation into themed insight predate user research. The affinity diagram comes from the Japanese anthropologist Jiro Kawakita, whose KJ method, developed in the 1960s, let researchers cluster field notes by emergent similarity rather than forcing them into categories chosen in advance. Themes surface from the bottom up. Contextual Design, from Hugh Beyer and Karen Holtzblatt in 1998, made affinity-driven synthesis a formal step between fieldwork and design, and the discipline tightened around triangulation, the principle that a finding is stronger when several independent sources point the same way.

The next shift was about storage. Tomer Sharon argued that the atomic unit of research should be a "nugget": a tagged observation backed by evidence🔎EvidenceValidationData supporting or refuting a hypothesisView reference →, stored so it can be searched and reused rather than buried in a slide deck nobody reopens. The practice that grew around this, often called atomic research and built into repositories like Dovetail, treats insights as immutable, evidence-linked facts that accumulate across studies. Research moved from a one-time report tied to a single project toward a queryable body of findings that compounds over time.

The naming converged too. Early schemas, including earlier versions of the Unified Product Graph, carried discipline-specific types such as ux_insight🔦InsightUser ResearchA synthesised finding from research and research_insight🔦InsightUser ResearchA synthesised finding from research. The field settled on a single, general notion of an insight that can come from a usability test, an analytics anomaly, a support ticket🆘Support TicketCustomer SuccessCustomer support request or issueView reference →, or a sales call, because the structure (an observation, the evidence, the why) holds whatever the source. Splitting it by discipline built artificial walls between findings that belonged together.

How it works in practice

A team runs eighteen onboarding interviews and tags four hundred observations. Affinity clustering surfaces a recurring shape: experienced users abandon the guided setup partway through and look for a way to skip it. That recurrence is a pattern. The team interprets it: these users trust their own ability to explore and read the wizard as condescension. That interpretation, backed by eleven observations across the eighteen sessions, is a finding with high confidence. The implication is stated plainly, offer an early escape to a blank workspace, which makes it actionable. Because it contradicts the prior belief that more guidance is always safer, it is marked surprising, which is the kind of insight that earns a roadmap🛣️RoadmapProduct SpecificationA strategic plan of features and milestonesView reference → slot. Stored as a node with its evidence linked, it stays queryable: two quarters on, a similar pattern elsewhere is recognised as the same cause rather than treated as new.

Insight vs. its neighbours

• Observation is the raw what, before the why. An insight is an observation plus its explanation and evidence. An unexplained observation is a fact with no instruction inside it.
• Evidence is a single data point with provenance; an insight is the synthesised meaning across many. One quote is evidence. The pattern eleven quotes form, and what it implies, is the insight. The graph builds the insight from evidence through observation_yields_insightObservationyieldsInsightcross-domain and insight_evidenced_by_quoteInsightevidenced byQuotehierarchy.
• Assumption🤔AssumptionStrategyA belief taken as true that underpins a strategyView reference → is an unvalidated belief about users. An insight is grounded in evidence. A proto-persona🎭PersonaUserAn archetype representing a user segmentView reference → or a roadmap bet may rest on assumptions; an insight is what the team has once the evidence is in.
• Hypothesis🧪HypothesisValidationA testable belief about a solutionView reference → points forward, stating that a change will cause an effect. An insight points back at what was learned. A confirmed hypothesis often produces an insight, which then seeds the next hypothesis.
• Opportunity🚪OpportunityDiscoveryA validated gap worth solvingView reference → is the actionable gap an insight reveals as worth pursuing. An insight explains something true about users now and informs an opportunity through insight_informs_opportunityInsightinformsOpportunitycross-domain, so the two stay separate nodes.

In the graph

In the Unified Product Graph, insight🔦InsightUser ResearchA synthesised finding from research is the canonical type that consolidates the older research_insight🔦InsightUser ResearchA synthesised finding from research and ux_insight🔦InsightUser ResearchA synthesised finding from research, in the Discovery, Research & Validation region, last in the research creation sequence after observations, quotes, and affinity clusters🧲Affinity ClusterUser ResearchA group of related observationsView reference →. That position is the point: an insight is the output of synthesis, never a freestanding claim. Inbound edges such as research_study_produces_insightResearch StudyproducesInsighthierarchy, observation_yields_insightObservationyieldsInsightcross-domain, and insight_evidenced_by_quoteInsightevidenced byQuotehierarchy enforce traceability back to the raw data. Outbound, an insight earns its keep by connecting forward: insight_informs_opportunityInsightinformsOpportunitycross-domain, insight_validates_needInsightvalidatesNeedcross-domain, insight_characterises_personaInsightcharacterisesPersonacross-domain, and insight_inspires_design_questionInsightinspiresDesign Questioncross-domain route it into strategy, users, and design. The insight_level property grades it from pattern to strategic, so a raw recurrence and a direction-changing finding are not filed as equals. An insight with no evidence behind it and nothing it informs is, structurally, an unsupported claim.