Conjoint analysis for pricing: how it works, when it's worth it, and cheaper options

Conjoint analysis is a survey method that shows people several versions of a product, each with a different mix of features and price, and asks them to choose. From thousands of those choices, a statistical model estimates how much each feature and each price level moves the decision. For pricing and packaging, it answers questions like "does adding priority support justify $20 more a month?" It is worth the effort when you have a real packaging decision, a few hundred reachable buyers, and time to design the study properly. Before that, cheaper methods usually get you further.

A worked example: packaging field-service software

Suppose you sell scheduling and invoicing software to commercial cleaning and maintenance companies. You are deciding what goes into two plans and what to charge for each. The open questions: does a mobile app for crews matter more than accounting integrations, and how much of a price gap will buyers accept between the plans? This example is hypothetical and used throughout the page.

How conjoint works

Sawtooth Software, which sells conjoint software, describes three steps:

1. Break the product into attributes and levels. In the example: price ($99, $149, $199, $249 or $299 a month), crew app (none, basic, full), accounting integration (none, one, several) and support (email, chat, phone).
2. Show respondents choice tasks. Each task shows a few product profiles built from different levels, and the respondent picks one. Sawtooth says a study usually has 8 to 20 of these tasks.
3. Model the choices. The model produces "part-worth utilities," a preference score for each level. Sawtooth notes that utilities can be compared within one attribute but not across attributes, and that its main deliverable is a market simulator: you set up competing offers and the simulator predicts shares of choice.

Wikipedia's overview says the method originated in mathematical psychology, credits Paul Green at Wharton with developing it for marketing, and credits Jordan Louviere with the choice-based approach. The earliest studies, in the 1970s, had people rank or rate profile cards instead of choosing.

For pricing, the key idea is that nobody is asked "what would you pay?" Sawtooth argues that question is hard to answer without context, while a choice between realistic offers at different prices lets you estimate price sensitivity indirectly.

How many respondents you need

This is where many startup conjoint plans run into trouble. Sawtooth's sample size rules of thumb give two quick checks:

• Start with 300 respondents, plus at least 200 per subgroup you want to report separately.
• Show every level at least 500 times across the whole sample, with 1,000 as the safer target. With c levels in the largest attribute, t tasks per person and a profiles per task, each person sees each level about t x a / c times.

In the example, price has 5 levels. With 10 tasks of 3 profiles each, one respondent sees each price level about 10 x 3 / 5 = 6 times. Reaching 500 exposures takes 500 / 6 = about 84 respondents; reaching 1,000 takes about 167. Sawtooth notes this second rule dates from older aggregate models and is only a ballpark. Its first rule, 300 respondents, is simpler, and Sawtooth says it works well in practice, so use it for planning.

Now compare that with the market. If you want separate results for small cleaning firms and large facilities companies, the rule suggests at least 400 qualified people. In a niche B2B market, finding 400 people who actually choose field-service software may be harder than running the analysis.

When conjoint is worth it

Conjoint fits when all of these are true:

• You are choosing between specific packages. You can list the attributes and realistic levels. If you still do not know which features matter, you are not ready.
• The decision is expensive to get wrong. A pricing change for an existing customer base, or a launch where you cannot easily test prices live.
• You can reach enough of the right buyers. Consumer products and large B2B categories usually can. A market with a few hundred buyers usually cannot.
• Someone can design it. Wikipedia notes that respondents simplify when shown too many attributes and that poorly designed studies can overvalue emotional features. Level ranges matter too: Sawtooth warns that attribute importance depends heavily on the range of levels you include.

Tool cost is often smaller than people expect. Sawtooth's pricing page lists a free plan for up to 50 respondents per survey that includes choice-based conjoint, with paid plans quoted on request and sold as annual subscriptions. The larger costs are recruiting qualified respondents and the design and analysis time.

Cheaper alternatives, and when to use them

Interviews about real purchases. Ask buyers to walk through their last comparable purchase, then show a specific offer and ask what would have to be true for them to approve it. This does not produce a demand curve, but it tells you which attributes belong in the study at all. How to research willingness to pay covers the method.

Best-worst ranking of features (MaxDiff). Respondents repeatedly pick the most and least important item from short lists. It ranks features for packaging with a simpler design than conjoint. Sawtooth's free plan includes it.

Direct price-sensitivity questions. Questions such as the Van Westendorp "too cheap / too expensive" set are quick to add to a survey. They measure stated reactions to price, not choices between packages, so treat them as a range to test.

Live tests with real offers. Quote different prices to comparable prospects, or test two plan pages. Real purchases beat any survey, but you need enough volume to tell the difference.

A sensible order for most startups: interviews to find the attributes that matter, a small MaxDiff or conjoint pilot under the free respondent cap to check your design, and a full study only when the decision justifies 300 or more qualified respondents. Quantitative market research covers survey sampling more broadly.

Your next step

Write down the packaging decision you are trying to make, the attributes and levels you would test, and how many qualified buyers you could realistically survey. If that number is well under 300, start with interviews.

To find those buyers, Instant Expert can search for people who match a description you write, such as "operations managers at commercial cleaning companies" or "product managers who priced B2B software plans." You review who it finds, it sends your invitations, and you pay only for calls that get booked. The directory pages for operations professionals in facilities management and product professionals in enterprise SaaS are one place to start.