MaxDiff vs conjoint: what each one is actually for

Both make people trade things off. Both produce scores. They are not substitutes, and choosing between them is easier than the literature makes it sound.

The one-line difference

MaxDiff ranks a list of separate things by importance. Conjoint predicts what people will choose between whole products.

MaxDiff asks: of these features, which matters most?

Conjoint asks: at £89 with next-day delivery and a two-year guarantee, versus £69 with standard delivery and one year, which do you buy?

MaxDiff, plainly

Also called Best-Worst scaling. People see a few items at a time, usually four or five, and pick the one that matters most and the one that matters least. That repeats over a handful of short screens.

Each screen is easy. Nobody has to hold twenty things in their head or interpret a scale. But because every item appears several times against different rivals, the forced choices add up to a ranking of the whole list, on a scale where the numbers are comparable.

What it is genuinely good at: taking a long list nobody can agree on and producing one clear order, with a line under the items that are true must-haves.

What it cannot do: tell you what somebody will pay, or how features interact. MaxDiff treats each item as independent. It does not know that free delivery matters far less when the product is cheap.

Conjoint, plainly

You define attributes and levels: price at three points, delivery at two, guarantee at two. The software builds product profiles from combinations of those levels and asks people to choose between them.

From the pattern of choices it works out how much each level contributes to a choice, which lets you simulate demand for products nobody has actually built, and estimate willingness to pay.

What it is genuinely good at: pricing, product configuration, and modelling a market where you need to know what happens if a competitor drops their price.

What it costs you: design complexity, a bigger sample, and a much heavier task for the participant. It also requires you to already know your attributes and their levels, which is a real constraint. If you are still working out what the attributes are, conjoint cannot help.

Choosing

If you need to	Use
Rank 20 features by importance	MaxDiff
Decide what goes in the next release	MaxDiff
Find which messages land hardest	MaxDiff
Set a price	Conjoint
Model a product against a competitor's	Conjoint
Decide which bundle to launch	Conjoint
Work out what your attributes even are	Neither, start with a card sort

The honest bit

SortedResearch runs MaxDiff. It does not run conjoint, and this article is not going to pretend otherwise.

If you need pricing research or a market simulator, you need a conjoint tool and probably somebody who has run one before. It is a heavier method and it deserves the respect.

What we would say is that a great many teams reach for conjoint when the actual question was "what matters most to our customers", and that question is answered better, faster and by far more cheaply with MaxDiff. The trade-off study is impressive. It is often impressive at answering a question nobody asked.

Where they meet

They pair well in sequence. MaxDiff first to cut a long list of candidate features down to the handful that genuinely move people. Conjoint second, on those survivors, to work out configuration and price.

Running conjoint on twenty attributes because nobody was willing to cut the list is how a study ends up needing thousands of respondents and producing a model nobody trusts.