The Ultimate Guide to Conjoint Analysis: Explanation, Examples, Types, Tools, & Use Cases
โConjoint analysis measures what matters to customers when they choose between products, by showing them competing profiles and recording which they pick. This is the hub guide: what it is, how a survey works, a worked example, and where to go for each part of the process in depth.โ
Conjoint analysis is a research method for understanding what's most important to customers when they're picking between products during a purchase decision.
People tend to call it an "advanced" research method. It isn't. Anyone able to use Google Forms can run a conjoint survey, and by the end of this page you'll know enough to design one.
This guide is the overview. Each section links to a full guide on that part.
What is conjoint analysis?
Conjoint analysis is a type of survey that measures how important customers think different attributes are, like price, brand or features, when comparing products in a purchase decision.
A conjoint survey shows two or more profiles that represent variations of the same product. Each profile contains the same list of attribute categories, but the options shown for those categories vary on each profile. A participant picks the profile they like most, that vote gets recorded, and a new set appears.
โฌ๏ธ interactive demo (via OpinionX) โฌ๏ธ
Those votes calculate two things at once: the importance of each attribute category (whether price or brand matters more when choosing a soda) and the ranking of options within each category (whether people pick Coca-Cola or Pepsi).
The four components of a conjoint survey
| Component | What it is | Called in OpinionX |
|---|---|---|
| Question | The context for the purchase decision being simulated | Question |
| Profile | A complete set of attribute values making up one potential product | Profile |
| Attributes | The characteristics a customer compares products on | Categories |
| Levels | The range of values inside each attribute | Options |
Academic writing says attributes and levels. OpinionX calls them categories and options, because those words need less explaining. Part-worth utilities are called scores for the same reason. Use whichever you're comfortable with.
A worked conjoint analysis example
You're running a conjoint survey to decide what burger to prepare for your weekend barbecue. Three categories: filling, sauce and bun type. Each has four options, so the filling could be pork, chicken, vegan or beef.
The survey answers two questions. Which categories drive the decision, meaning whether the filling matters more than the sauce or the sauce more than the bun? And which options win inside each category, so would switching the pork patty to chicken, beef or vegan make more people want your burger?
Example conjoint analysis results, showing relative Importance of categories/attributes (left) and options/levels (right)
Our example results show that 50% of the decision came down to the filling, while bun type accounted for only 10%. Inside the filling category, beef and chicken were the two most popular.
That's conjoint in one paragraph: a percentage on each category, and a ranking inside each one.
How do you design a conjoint survey?
Three configuration decisions carry most of the weight.
Profiles per set is how many appear on screen at once. Anything between 2 and 6 works, and the more complex your categories, the lower it should go. Sets per participant is how many times each person votes before moving on, where eight is a reasonable default. Then there's the skip button: leave it in and participants can decline a set they find irrelevant, take it out and you get forced voting on every set.
Most conjoint surveys also carry three other question types alongside: an identifier question to know who responded, multiple-choice questions to make segmentation possible later, and a text response to catch categories you missed.
For the mechanics behind those numbers, see how the results are calculated. Mobile conjoint covers designing profiles that survive a phone screen.
How do you read the results?
The score for an option is the likelihood that a profile gets picked when that option appears on it. The scale runs from -100, meaning certain not to be picked, to +100, meaning certain to be picked.
The score for a category is the gap between its highest and lowest option scores. If iPhone scores 27 and Huawei scores -56, the Brand category scores 83.
That distinction matters. Option scores show relative preference, while category scores show how much influence a category has over the decision, which isn't the same as it being good news. A large category score can mean one option inside it's disliked intensely, so always read the option scores underneath.
You can also combine option scores to predict which of two hypothetical profiles wins. Pair the top-ranked brand with the worst memory and price, put it against the worst brand with the best memory and price, and the scores tell you which profile a customer picks.
The workings behind all of that are in how the scores are calculated, and the market simulator models how a change to your product moves market share.
How do you segment the results?
This is the part that earns the survey its keep. You aren't looking for which attributes matter to customers, you're looking for which attributes matter to your most important customers.
Clicking any bar on a multiple-choice or rating-scale result recalculates the whole results page for that group. Two segments can also be put side by side: in one survey, iPhone was the top option for current iOS users and 10th of 15 for Android users, which was the largest gap between the two segments on any option. The Segments tab goes further again, laying out a colour-coded crosstab of every segment at once, up to 20 groups in one table.
The method behind it is needs-based segmentation, and crosstab analysis covers how to read the tables once you have them.
What are the types of conjoint analysis?
"Conjoint analysis" is a catch-all covering a lot of different survey formats. The main ones:
Choice-Based Conjoint (CBC) is the standard. Participants see 2 to 6 profiles and pick their preference, repeatedly, with the levels changing each round.
Best-Worst Conjoint asks for the best and worst profile in each set, collecting more data per round at the cost of more effort.
Ranking-Based and Constant Sum conjoint replace the single pick with a full ranking or a points allocation across the profiles.
Adaptive Choice-Based Conjoint (ACBC) learns from earlier answers and changes what it shows next. It takes longer, costs more, and generally needs expert support.
Menu-Based Conjoint (MBC) lets participants build their own product from a menu, which is the closest format to a real configurator.
Then there are two that aren't survey formats at all. Time-Series Conjoint analyses existing results in waves by completion date. Latent Class Conjoint finds groups with similar preferences inside the data, which is segmentation by another name.
Each format gets an example in 13 types of conjoint analysis, and full-profile conjoint has its own guide.
What are the best tools?
The market runs from free self-service tools to platforms costing tens of thousands a year that expect you to hire a consultant alongside them.
Most conjoint software treats the method as a gated enterprise feature. Several of the largest survey platforms either put it behind their top tier, sell it as a managed service you can't run yourself, or charge a separate add-on on top of an existing licence.
Ten platforms are reviewed with current pricing, screenshots and verdicts in the tools guide, including the ones worth choosing over us.
When should you use it?
Research that simulates a purchase decision is the right fit, which is why conjoint shows up so often in pricing research
A conjoint survey works best when all four of these hold:
You're simulating a straightforward, considered purchase between similar products
One person makes the purchase decision
The customer already knows what kind of product they need
You know which attributes the customer compares products on
Miss any of those and another method will probably serve you better.
Three projects conjoint suits
Willingness to pay. Customers often don't know their own maximum price, and wouldn't tell you if they did. Conjoint sidesteps the question by watching which profiles get chosen. Include price as a category and you can convert score differences into currency.
Market share modelling. Utility scores predict how market share would divide across a set of competing products, and how changing a feature moves that split.
Product bundling. TURF analysis finds which combination of options reaches the largest audience, which is a different question from bundling the two highest-scoring options. If your top two are loved by the same people, bundling them reaches nobody new.
The failure cases are covered in detail in when not to use conjoint, and marginal willingness to pay works the price conversion through with worked numbers.
What are the alternatives?
Most projects people consider conjoint for don't need it. Eight methods cover the ground more cheaply and with far less setup.
Pairwise comparison, points allocation, ranked choice voting and MaxDiff analysis are all choice-based, so they force the same trade-offs conjoint does without the profile structure.
Agreement voting, the Kano Model, Van Westendorp and TURF analysis aren't choice-based. Van Westendorp in particular is the one to reach for when you have no price benchmark at all, since conjoint needs 2 to 7 candidate prices before you can start.
All eight are compared with advantages, disadvantages and suggested tools in alternatives to conjoint analysis. There's also a dedicated comparison against the Analytic Hierarchy Process.
Running conjoint on OpinionX
Conjoint on OpinionX is the Conjoint Rank block. Setting one up means adding your categories and their options, then configuring profiles per set, sets per participant and whether the skip button stays.
OpinionX builds profiles to work on a phone. They reshape to the screen, stack horizontally or vertically, and put category titles above each option so nobody has to scroll back and forth to work out what they're looking at.
Every question type and every analysis feature is unlocked on the free tier, capped at 25 participants per survey. That's enough to run a full conjoint study, read the segmented results and validate your design before paying anything. The Analyze plan is $900 a year and removes the participant cap. Accelerate is custom-priced for enterprises needing SSO, MFA and procurement support.
Frequently asked questions
What is conjoint analysis in simple terms?
A survey that shows people competing versions of a product and asks which they'd pick. Repeat that across different combinations and the pattern of choices reveals which attributes drove the decision and which options within them people preferred.
What are attributes and levels?
Attributes are the characteristics being compared, like brand, storage and price. Levels are the specific values inside each one, like iPhone, 256GB and $899. Each profile takes one level from every attribute. OpinionX calls them categories and options.
What is a part-worth utility?
The score for one option, expressed as how much it raises or lowers the chance of a profile being chosen. It runs from -100 to +100, and it only means anything relative to the other options in the same category.
How many participants does a conjoint survey need?
It depends on your configuration instead of a flat number: how many categories and options you have, how many profiles per set, and how many sets each person completes. More categories and options need more data to separate.
Is it hard to run?
No, and the reputation comes from the pricing instead of the method. Setting up a conjoint survey means listing your categories, listing the options inside each, and choosing three configuration numbers. Reading the results means understanding one scale from -100 to +100.
Over 42,000 researchers and product people get one method breakdown like this each week in The Full-Stack Researcher.
About The Author:
Daniel Kyne is the Co-Founder of OpinionX, a free research tool for stack ranking peopleโs priorities โ used by thousands of product teams to better understand what matters most to their customers. OpinionX has a bunch of free research methods for ranking peopleโs preferences โ including free Conjoint Analysis surveys alongside other ranking methods like Pairwise Comparison and Points-Based Voting.
โ Create a FREE Conjoint Analysis Survey now