Conjoint analysis in marketing: how it works, the design choices that matter, and a worked example
Conjoint shows what buyers trade off when price, features and brand compete, by making them choose rather than rate.
Published by AI Knows Us · Updated 12 October 2026
Conjoint analysis is a survey method that measures how much each feature of a product matters to buyers, by asking them to choose between realistic product versions. Each version mixes features and a price. From the pattern of choices, the analysis calculates a value for every feature level, which marketers call a part worth or utility. The result answers questions such as "how much more will customers pay for the longer warranty?" and "which two features should the entry model have?" It is the standard method for pricing and feature trade off decisions where asking buyers directly would produce wishful answers.
Why asking "what matters to you" does not work
Ask a buyer to rate the importance of price, quality, delivery speed and brand, and every item comes back as important. Nothing in the question forces a trade off. In a real purchase, the buyer cannot have the lowest price and the fastest delivery. Conjoint recreates that constraint: each choice task shows two to four product versions, and the respondent picks the one they would buy. The analysis learns from what people give up, not from what they say they want.
The vocabulary
| Term | Meaning | Example for a project management app |
|---|---|---|
| Attribute | A feature that varies between versions | Price, storage, integrations |
| Level | A value an attribute can take | Price: a low, a middle and a high monthly fee per user (levels you choose) |
| Profile | One product version made of one level per attribute | Middle price, large storage, ten integrations |
| Choice task | A screen showing several profiles, one to be chosen | Three profiles and a "none of these" option |
| Part worth | The value the analysis assigns to a level | The pull of large storage against small storage |
| Importance | Share of the total value that an attribute explains | Price carries a larger share than storage |
The main types
Choice based conjoint
The most used form. Respondents see sets of profiles and pick one, or none. It mirrors a shelf or a pricing page and works well for pricing. This is the default for most marketing studies.
Full profile and ratings based conjoint
Respondents rate or rank individual profiles. It is older and simpler, and it limits the number of attributes a respondent can handle.
Adaptive and menu based conjoint
Adaptive versions change tasks as the respondent answers, which helps with many attributes. Menu based versions let respondents assemble a package from options, which suits bundles and add ons.
A worked example
A business sells a booking tool to small salons. It wants to decide which plan to launch first. The attributes and levels below are invented for illustration.
| Attribute | Levels |
|---|---|
| Monthly price | Low, middle, high (three set prices) |
| Text reminders | Not included, up to a set number, unlimited |
| Online payments | No, yes |
| Staff calendars | One, up to five, unlimited |
| Support | Email, email and phone |
Each respondent, a salon owner, sees a series of screens, each with three plans built from these levels plus "none of these". After enough screens, the analysis reports something like this: unlimited reminders add more pull than phone support; the move from the middle to the high price lowers choice sharply; online payments are close to a requirement. The business can then build a plan with the features that carry the most value per unit of cost and set a price near the point where choice starts to fall. The pattern is the useful output, the exact numbers depend on the study.
The design choices that decide quality
- Pick attributes that buyers recognise and that you can change. A feature you cannot build is a wasted attribute.
- Keep the list short. Respondents tire when each screen has too many attributes. Five to seven is a common range, and fewer is easier.
- Choose realistic levels. Price levels must straddle what the market charges, including one above and one below your plan.
- Include a "none" option. It shows whether a buyer would walk away, which is what happens in real markets.
- Sample the buyers you want. Current customers, lapsed customers, and buyers of a rival can all matter. Screen so that every respondent can actually buy.
- Test the survey. Run it with a handful of people first, and watch where they hesitate or misread.
What you get out and what to do with it
- Part worths and importance scores that rank the attributes.
- Willingness to pay for a feature, estimated from the trade off against price.
- A simulator. A spreadsheet or tool that predicts the share of choices for any plan you build, including a rival's plan.
- Segments. Groups of buyers with different priorities, such as price sensitive and feature led.
The simulator turns the study into a decision tool. Enter your plan and each rival plan, and read the predicted share of choices. Try changes one at a time, and write the result next to the decision it supports. If you need background on writing the question and the brief for a study, marketing research explained covers the process, and perceptual mapping is the matching method for how buyers see brands rather than how they trade off features.
Tools and suppliers
Conjoint can be run with survey software that has a conjoint module, with specialist conjoint software, or through a research agency. Sawtooth Software, Conjointly and Qualtrics each publish conjoint tools, and you should check each site for current plans and features before choosing. A research agency adds design, sample and interpretation. For help judging an agency's method, read how to judge a B2B market research agency.
Limits to plan around
- It measures stated choices. Check the results against sales data where you have it, and treat the study as one input.
- It prices only the attributes you include. A feature left out cannot be valued.
- It suits new or changing offers. For a stable product with long sales history, the sales data may answer the question more cheaply.
- It needs a sample large enough for the segments you want to read. Decide the segments before you set the sample.
Plan these in the brief and the study works as designed.
When the buyer's choice is made inside an assistant
Conjoint asks what buyers trade off when they see your plan next to a rival's. Today a growing share of that comparison happens before they see either plan: they ask an assistant which option suits a salon of five and the answer lists features, prices and brands in a few sentences. The features that the assistant names are the ones your pages state plainly. AI Knows Us (aiknowsus.com) asks the assistants those comparison questions and records which brands and which features come back, so you can set the attributes your conjoint study says buyers value beside the ones the assistants actually repeat.
Questions about conjoint analysis in marketing
What is conjoint analysis used for?
Pricing, feature selection, packaging of plans, and forecasting the share a new offer would win against rivals.
How many respondents do I need?
It depends on the number of attributes and the segments you want to read. Ask the supplier to explain the sample size for your design, and decide your segments first.
What is the difference between conjoint and a normal survey?
A normal survey asks people what they think. Conjoint makes them choose between products, so the data shows trade offs.
Can a small business run conjoint?
Yes, with a tool that has a conjoint module and a sample of reachable buyers. Keep the design small, with a few attributes and levels.
Is conjoint the same as A/B testing?
No. A/B tests compare real versions with live traffic. Conjoint tests many hypothetical versions with a survey sample, before anything is built.