An average sentiment score of 3.8 out of 5 tells you almost nothing. The individual reactions it summarises tell you everything.
Product research has a chronic averaging problem. We take the diverse, contradictory, specific responses of real people and compress them into a single number that loses precisely the information we need.
What averages destroy
Imagine a feature evaluation: two strongly positive responses, one conditional, one mixed, one strongly negative. Average score: 3.2. Interpretation: broadly lukewarm.
But the strongly negative response is from your enterprise buyer archetype, citing a specific security concern. The conditional response is from your power user, who would adopt it immediately if one edge case were handled differently. The mixed response is from a user who misunderstood the feature.
The average hides the signal. The individual responses contain the roadmap.
Per-persona research
Swarm-Lite was built around this insight. The output of a Huddle session is not an aggregate score — it is a set of per-persona responses, each with specific objections, conditions, and sentiment. The CTO archetype’s security concern is preserved. The power user’s edge case is surfaced.
This is harder to present than a bar chart. It requires the reader to do synthesis. But synthesis is the product manager’s job — and the inputs to that synthesis are what the research should provide.
The practical implication
When you design a research system, decide whether you are trying to confirm a hypothesis or discover a blind spot. Averages confirm. Personas discover.
Design for discovery.