A list of highly rated features can leave a team almost where it started. Customers consider everything useful, while the budget still covers only a fraction of the work. MaxDiff introduces a more demanding question: among these few options, which matters most and which matters least?

Participants answer that question across a series of sets drawn from a longer list. The pattern of choices is then analysed to estimate relative preferences. Also known as best–worst scaling in this common item-comparison form, MaxDiff is described in Sawtooth Software’s method comparison. It is often used for benefits, messages or features that would be cumbersome to rank all at once.

Decide what “best” means

The criterion needs to remain clear throughout the exercise. “Most important when choosing a provider” is different from “most appealing” or “most believable”. Combining those judgements in one instruction leaves the resulting scores difficult to interpret.

Items should also be comparable. An illustrative study of account-management priorities might include clearer invoices, faster refunds and easier permission management. Adding “a service you can trust” mixes a broad aspiration with specific capabilities, making the comparison less informative. Research the language and scope of the items before constructing the sets.

Use a planned design that gives items suitable exposure and connections across the study. It is not necessary for every respondent to see every possible pair, but casually selecting sets can produce uneven information. Pilot the whole task sequence, since choosing from one small set may be straightforward while repeating the exercise becomes tiring.

What the ranking leaves unanswered

A relatively low score does not establish that an item is unwanted. It may be less important than the other items shown, or an expected basic requirement that respondents take for granted. The composition of the list matters, as does the audience. Segment differences may be more useful than a single overall ranking.

Interpret the numbers according to the scoring method. Some analyses report rescaled preference shares and others use different utility scales; a score of 20 does not automatically mean something is twice as important as an item scoring 10. Ask what comparisons the model supports before presenting ratios as facts.

MaxDiff can inform feature prioritisation, but the team still needs to consider effort, dependencies and evidence of the underlying problem. If the decision concerns bundles of features at different prices, conjoint analysis may fit better. MaxDiff compares items; it does not, by itself, establish which complete offer someone would buy.