Conjoint analysis is a survey-based statistical method for measuring how customers value different attributes of a product or service. It is the gold-standard technique for pricing, feature prioritisation and portfolio design.
How it works
Rather than asking customers directly how much they value an attribute — which produces unreliable, inflated answers — conjoint presents them with a series of realistic trade-offs between concepts that vary across attributes (e.g. brand, price, speed, warranty). From their revealed choices, the method mathematically extracts the implicit utility they place on each attribute level.
The core model is based on random utility theory. Each respondent is assumed to have an underlying utility function:
$$U_{nj} = \mathbf{x}_j^{\top}\boldsymbol{\beta}_n + \varepsilon_{nj}$$
where $\mathbf{x}_j$ is the attribute vector for alternative $j$, $\boldsymbol{\beta}_n$ is respondent $n$'s part-worth utilities, and $\varepsilon$ is a random error. Choice probabilities follow from the multinomial logit kernel.
What it tells you
From the estimated utilities, three decision-ready outputs emerge:
Attribute importance — which features drive choice, ranked
Willingness-to-pay — the money-denominated value of each feature level
Market share simulation — forecasted share-of-preference for any product configuration in any competitive context
Varieties of conjoint
CBC (Choice-Based Conjoint) — the standard. Respondents pick one from a set of alternatives.
ACBC (Adaptive CBC) — the design adapts to each respondent via Build-Your-Own, screener and tournament stages. Better for products with many attributes.
MaxDiff — best-worst scaling for prioritising long lists of features or messages.
MBC (Menu-Based Choice) — for products where customers assemble a basket rather than pick one.
History
Conjoint analysis was introduced in 1971 by Paul Green and Vithala Rao at Wharton, drawing on mathematical psychology (Luce) and econometrics (McFadden’s random utility theory, for which he later won the Nobel). The modern Hierarchical Bayesian estimation framework became standard in the late 1990s with the work of Allenby, Rossi and McCulloch, enabling individual-level utilities even on sparse per-respondent data.
When to use conjoint
Pricing a product in a competitive market where you need to know your optimal position
Prioritising which features to build or which to cut from a crowded roadmap
Portfolio design — choosing which SKU variants to launch and which to retire
Market sizing — estimating how a new product will share from existing competitors
Segmentation — discovering distinct preference groups you can target differently
If you can describe your decision as “which bundle of attributes should we offer, and at what price?”, conjoint is the right tool.
Related terms
Hierarchical Bayes · MaxDiff · Menu-Based Choice (MBC) · Willingness to pay · Full methodology