Insights

Vectors don't understand meaning. They map distance. That limitation is exactly why they make an honest backbone for qualitative work at social media scale.

Qual has always had a cherry-picking problem. Vector maths gives it a map of the whole discourse, so saturation becomes something we can show rather than assert.

Our idle servers used five to ten times more energy doing nothing than the Reading Machine used reading everything. So we changed the architecture.

Three weeks before the Hungarian vote we called a two-thirds supermajority from 10,000 conversations while polls spread across 14 points. The election was the proof, not the point.

TensionScope identifies the deep psychological frictions — the Tensions — that actually drive consumer decision-making.

At QUAL360 in Washington, 57% of talks used social as an insight source. Almost none of it was rigorous. Mediocre at scale is still just mediocre.

Is asking the best way to understand? At the ESOMAR Citizen Insights Summit in Brussels, I argued that institutions should listen to citizens before they survey them.

When a survey's first question has no option that speaks for you, the data is skewed before it's analysed. A story about the ESOMAR CEO survey, and what happened after I complained.

"Social isn't representative, people just like to complain." True, and beneath every complaint lies a story about expectations. All you need to do is invert the comment.

Notes from IIeX North America: the more data you digest, the better your insights get, went the pitch. My worry is that our tools are outshining research's primary mission.