Research leadership · point of view
The bottleneck is moving from data to judgment.
Twenty years building research organizations — and the last three spent working out which parts of research an AI can be trusted to do.
For thirty years research was constrained by collection. Fielding was slow and expensive, so methodological skill largely meant knowing how to get an answer at all. AI collapses that constraint — and what becomes scarce is not data but judgment: which questions are worth asking, and when a generated answer is trustworthy enough to bet a launch on.
Ground truth
Human respondents, real sample, known error. Slow, costly, and the only thing a synthetic estimate can be checked against.
Modeled
Synthetic samples and machine synthesis. Fast and cheap — and only as good as the boundary someone drew around where it holds.
Research will not shrink because of this. It expands: cheaper research means far more of it. The risk is a flood of confident, unvalidated insight, and the first expensive wrong answer will make buyers demand provenance. The durable advantage belongs to whoever can say precisely why an answer should be believed.
Two AI transformations, from two different chairs.
Inside an enterprise insights function · Tyson Foods
Introduced AI-enabled synthetic consumer panels and AI-based packaging research into a multi-business-unit insights organization on a $4.5M annual research budget. Validated before trusting: ran synthetic samples against ground-truth human data, established which question types held up and which did not, and wrote that boundary into the standard. Roughly halved innovation cycle time — but the reason leadership acted on the output instead of debating it was the validation work, not the speed.
As the owner, carrying the P&L · Alloy Personal Training
Co-owned a health and fitness business across three franchise territories and rebuilt its operating spine on automation — lead capture, nurture, booking, membership and follow-up as one system — with agentic workflows on top for the recurring work. Scoped it, built it, and lived with what it produced, including payroll. The automation is a meaningful part of why the business sold on the terms it did.
What I can actually run myself.
Quantitative
- Segmentation
- MaxDiff · TURF
- Driver analysis · regression
- Significance testing · weighting
- Conjoint · demand spaces
- Predictive modeling
Qualitative & mixed
- Jobs-to-be-Done
- Ethnography · IDIs
- Concept & claims testing
- Journey mapping
- Foresight · trend synthesis
Sensory
- Descriptive analysis
- Duo-trio · tetrad
- Response surface methodology
- Shelf-life modeling
- CLT · HUT design
Tools & systems
- SPSS · Excel · Tableau
- Fairgen synthetic panels
- Behaviorally
- Nielsen · IRI · NPD
- API automation · agentic workflows