Four groups, found from what moves them
Not age or income. The engine modelled what drives each buyer's loyalty, grouped people whose drivers look alike, and let the data say how many groups there are. Pick a persona to see its full profile.
What buyers say matters runs opposite to what drives their loyalty
Twelve ideas in, five to the full test
Each buyer rated four of the twelve concepts. The grid shows the share who would buy, by persona, with the base and a 95% interval. A gold box means that persona's lead is clear of every other persona's interval.
Select a concept name for its description and what buyers said about it.
Pick the top five overall and one persona is left with almost nothing
Build a range, see who it reaches
Choose up to five concepts. Reach is the share of buyers who would definitely buy at least one of them, persona by persona.
Definitely buy at least one
Each persona's best option in the range
Each buyer rated four concepts, not all twelve, so reach for a combination is modelled: within each persona, the chance of definitely buying at least one concept, treating concepts as independent, then weighted by persona size. The best-option figures are measured directly.
What to say, and what they will pay
Buyers chose the most and least appealing claim from six sets of four. Price comes from four questions about Ferrow's core 340 g bag, in US dollars.
Best minus worst, from -100 to +100
Acceptable range and optimal price, 340 g bag
Ask them directly
Each persona answers in the first person, from the study. Every answer is labelled: Measured comes straight from the data, Modelled from the driver models, Inferred goes beyond the data and says so, and Declined means the study has no evidence for that group.
This public demo is scripted. The live engine answers your own questions on your own data.
Same idea, four ways to say it
Each persona carries a measured motivational profile: what tends to drive the decision in the moment, not a fixed personality. The engine writes through it, so a persona answers, and a message lands, in the register it responds to.
Does the answer hold in the moment of choice?
People do not decide in the state they answer surveys in. A rush, a stock-out or a price rise can change what someone wants from a coffee. Here each concept is read twice: at rest, and with each buyer's dominant motivation under pressure.
A stress test, not a forecast. It shows which ideas depend on buyers staying in their usual frame of mind, and which ones hold up when they are not.
We hid four personas in the data. Did the engine find them?
Because this study is simulated, the truth is known: who belongs where, what drives each group, and which concepts each group prefers. The engine never saw it. These are its answers checked against that truth.
Predicting reactions of 400 buyers the model never saw
The data chose four
The elbow is the point where adding another group stops buying much. Nobody told the engine there were four.
Personas found against personas planted
Planted drivers against drivers found, top three per persona
The engine recovers structure that is known to be there, from what drives loyalty, and personas built that way predict the reactions of people it has not seen far better than demographic groups. The same checks run on client data, where the truth is not known: a held-out test, persona separation and answer labelling.
Accuracy on a real market: real data is noisier. Range reach is modelled. Under pressure is a stress test. Tested on real people as well: in a public dataset of 2,058 Americans (Twin-2K-500, grocery purchase intent), driver personas explained about three times as much of the difference between held-out respondents as demographic groups did, 20.3% against 6.5%. Read the public-data test →
Retrofit
Turn a segmentation you already have into personas you can question, when the study holds respondent-level data.
Full build
A new study designed for it: the outcome, the drivers, a short trade-off exercise, and the personas built from the start.
Upgrade
Add the motivational lens, concept reads and the checks to personas you already use.