Conjoint Analysis

What it is

A trade-off-based survey method that infers how much buyers value individual product features and price levels, without asking them directly what they'd pay for each one. Respondents are shown a series of product "bundles", each a different combination of features and price, and simply pick which bundle they'd choose. Because each bundle forces a trade-off (more storage but a higher price; SSO included but a longer contract), the pattern of choices across many bundles lets statistical modelling back out a value, in price terms, for every individual feature. Where Van Westendorp answers "what should this whole product cost" and Value Metric/WTP answers "what should we charge for", conjoint analysis answers a narrower, more rigorous question: how much is each individual feature worth to each persona, in isolation from the others. It is the most statistically rigorous of the three pricing methods in this category, and the most resource-intensive; running it well is a project, not a quick survey.

When to use it

• Bundling or unbundling a major feature (moving SSO, advanced reporting, or an integration from one tier to another) and you need to know its standalone price value, not just a gut feel
• Building a new product from scratch with many candidate features and no existing price to anchor against
• Van Westendorp or Value Metric/WTP research has produced a price range or a chosen value metric, but you now need to decide exactly which features justify moving a customer up a tier
• Facing an internal dispute about which of several roadmap features should be "premium-only", with genuine disagreement about which one buyers would actually pay more for
• Preparing a major re-platform or packaging overhaul where the cost of getting individual feature pricing wrong (a full sales-team retrain, a customer remigration) is high enough to justify the extra research rigour
• You have the budget and respondent access (typically 150+ per segment) that a smaller, faster method like Van Westendorp does not require

How to run it

1. Select the features and price levels to test. List 4–6 features or attributes that are genuinely still in question, plus 3–4 price levels per feature. More than 6 attributes overwhelms respondents and degrades data quality; if you have more candidates, run a smaller pre-study (a simple ranking exercise) to cut the list down first.
2. Choose a conjoint design. For most PMM pricing questions, a "choice-based conjoint" (CBC) is the right design: respondents repeatedly pick their preferred bundle from a small set (usually 3, including a "none of these" option) rather than ranking or rating every combination. CBC mirrors how buyers actually decide and produces more reliable data than older full-profile ranking methods.
3. Generate the bundle sets. Use conjoint software (Sawtooth, Qualtrics Conjoint, or a comparable tool) to generate a statistically balanced set of bundles, so each feature and price level appears often enough, and in enough combinations, for the model to isolate its individual effect. Do not hand-build bundles; a non-randomised design confounds the features with each other and the resulting estimates are unusable.
4. Field the survey to a representative sample. Recruit at least 150 respondents per persona or segment you plan to analyse separately; each respondent typically answers 8–12 choice tasks. Screen respondents to match your actual buyer profile exactly as you would for any other pricing survey; a mismatched sample produces confident-looking numbers for the wrong population.
5. Run the choice tasks. Present each respondent with their assigned sequence of bundle choices. Keep the survey to 15–20 minutes; fatigue past that point measurably degrades response quality, particularly in the final few tasks.
6. Estimate part-worth utilities. Run the statistical model (hierarchical Bayes estimation is standard) to produce a "part-worth" score for every feature and price level tested, for each respondent. Converting a part-worth into an actual currency value requires anchoring it against the price attribute in the same model.
7. Convert part-worths into willingness-to-pay figures. For each feature, calculate the price increase a typical respondent in a segment would tolerate in exchange for adding that feature, holding all else constant. This is the number that answers "what is SSO worth" in pounds, not just in relative preference.
8. Run a market simulation. Simulate preference share across two or three candidate packaging scenarios (for example, "SSO in Growth tier" versus "SSO in Enterprise tier only") before committing to either, since the simulator shows which allocation maximises overall preference and revenue against the tested price points.
9. Feed results into packaging and pricing. Hand the per-feature value estimates to whoever owns the Good-Better-Best tier structure, so premium features are fenced into the tier that best matches the segment willing to pay for them, and price the ladder using willingness-to-pay figures rather than a rounded guess.

Cadence & ownership

PMM typically commissions and owns interpretation of a conjoint study, partnering with a research vendor, an in-house research/insights team, or a data science partner to handle the statistical modelling; few PMMs run the hierarchical Bayes estimation themselves. This is not a continuous programme like Voice of the Customer; run it as a discrete project ahead of a major packaging or pricing decision, and re-run it only when the feature set or competitive landscape has shifted enough to make the old part-worths unreliable, typically every 18–24 months, or immediately ahead of a major re-platform or bundling change. Budget 6–10 weeks end to end (design, fielding, analysis) and a meaningfully larger budget than Van Westendorp or Value Metric/WTP, since sample size requirements and vendor/software costs are both higher.