Find what your customers value most
Show people four or five items at a time and ask which matters most and which matters least. MaxDiff turns those picks into a ranking on one scale, and Jean adds the reasons behind it.
- 1Scans receipts22Best 48%Worst 4%
- 2Price alerts18Best 39%Worst 6%
- 3Shared lists with family14Best 31%Worst 11%
- 4Meal plans from what you bought12Best 27%Worst 14%Above the line: would make them switch apps
- 5Coupons applied at checkout11Best 25%Worst 15%
- 6Store map with aisles9Best 18%Worst 22%
- 7Recipe videos8Best 15%Worst 26%
- 8Works offline6Best 10%Worst 38%
Scores sum to 100. Scans receipts is about twice as preferred as coupons.
Run it with
Methods
Preference scores
Each item gets a score from Hierarchical Bayes, scaled so all items sum to 100, so a 22 is about twice as preferred as an 11.
Best, worst and net counts
How often each item was picked best and picked worst when shown, which shows the items people love and the ones they reject.
Anchored MaxDiff
A follow-up question draws a line through the ranking, so you know which items would actually change a purchase, not just which beat the others.
TURF
Finds the set of three or four claims or features that reaches the most people, so you build a lineup instead of picking one winner.
Segment rankings
The same scores split by segment, or by groups found in the data, so you see where the order changes.
A ranking, not a tie
Rating scales let people call everything important. MaxDiff forces a trade-off in every set, so the gaps between items are real and comparable.
Forty items in minutes
A thousand digital twins work through a balanced design of 12 to 15 sets each. Rank a long list of features, claims or benefits and cut the bottom half before you pay for a real sample.
Hear why it won
The AI interviewer takes real customers through the same sets and asks why they picked what they did. You get the score, the line that matters, and the reason in their words.
How you run it
- 1Minutes
Screen
A thousand digital twins answer first, so the weak options drop out.
- 2Overnight
Interview
The AI interviewer talks to real people and follows up on what they say.
- 3By morning
Decide
The number, the reasons behind it and the quotes that carry it.
Questions
What is MaxDiff analysis?
MaxDiff, also called best-worst scaling, shows people small sets of items and asks which is most and least important. Across 12 to 15 sets, the picks give every item a preference score on one scale.
MaxDiff vs. a rating scale: when should I use it?
Use MaxDiff when you need to rank many items and most of them sound good. Ratings let everything score high, and people use scales differently across cultures. MaxDiff makes people choose, which separates the items.
How many items, sets and respondents do I need?
Fifteen to forty items is typical. Each person sees four or five items per set, and each item should appear about three times per person. For real samples, plan on 200 or more per segment you want to read.
What is anchored MaxDiff?
Standard MaxDiff tells you the order, not whether any item matters. Anchored MaxDiff adds a question on which items would change a decision, which places a threshold in the ranking.
MaxDiff vs. conjoint: which do I need?
MaxDiff ranks a list of separate items, such as claims or features. Conjoint tests full products built from attributes and levels, often with price. Many teams run MaxDiff first to choose what goes into the conjoint.
Can digital twins replace a MaxDiff with real people?
Use twins to rank a long list quickly and cut the bottom of it. Confirm the top items with real customers before you commit budget to them.