How to run a conjoint analysis study
A practical walkthrough of choice-based conjoint analysis: how to define attributes and levels, design choice tasks, size the sample, and turn part-worth utilities into pricing and packaging decisions.
How to run a conjoint analysis study
Conjoint analysis is a survey-based method that measures how buyers trade off product attributes, such as price, features, and brand, by asking them to choose between realistic product profiles instead of rating features in isolation. Run it by defining a handful of attributes and their levels, generating a set of choice tasks, collecting responses from a representative sample of real buyers, and then reading the part-worth utilities to see what drives preference and what people will pay for.
This guide walks through the full workflow: what conjoint is, the common types, how to define attributes and levels, how to design choice tasks and size the sample, how to interpret the output, and how to turn it into pricing and packaging decisions. It closes with how conjoint differs from MaxDiff and how to recruit the buyers who make the numbers trustworthy.
What conjoint analysis actually measures
The core idea is simple: people reveal what they value not by telling you, but by choosing. When a buyer picks one product over another, they are making a trade-off between everything that product offers at the price it costs. Conjoint analysis reconstructs the hidden math behind those choices.
Instead of asking “how important is price on a scale of 1 to 5,” which almost everyone answers “very important,” conjoint shows buyers competing product profiles and forces a decision. Repeated across many choices and many respondents, the pattern of decisions lets you estimate a numeric value, called a part-worth utility, for every level of every attribute. Those utilities are the raw material for pricing, packaging, and feature prioritization.
Conjoint sits alongside other structured trade-off and prioritization methods. If you are still choosing an approach, the complete walkthrough to product research methods, frameworks, and best practices puts conjoint in context with interviews, surveys, and usability testing, and the market research methodology guide covers when quantitative trade-off methods are the right tool at all.
Common types of conjoint analysis
There is no single conjoint method. The main variants differ in how they present choices to respondents.
- Choice-based conjoint (CBC) is the most widely used today. Respondents see a small set of full product profiles and pick the one they would choose, often with a “none of these” option. It closely mimics a real buying decision and handles price well, which is why most pricing studies use it.
- Adaptive conjoint analysis (ACA) adjusts the questions based on earlier answers, focusing on the attributes that matter most to each respondent. It suits studies with many attributes but is more complex to build.
- Ratings-based (traditional) conjoint asks respondents to rate or rank individual profiles rather than choose between them. It is older and less common now because rating in isolation is a weaker signal than choosing.
For most teams running pricing, feature, or packaging research, choice-based conjoint is the default, and the rest of this guide focuses on it.
Step 1: Define your attributes and levels
Attributes are the dimensions of your product that you want to test. Levels are the specific options within each attribute. Getting this structure right is the single most important part of the study, because everything downstream depends on it.
A few rules keep the design clean:
- Keep it to four to six attributes. More than that and respondents start taking mental shortcuts, which adds noise rather than insight.
- Use two to five levels per attribute. Levels should be realistic, mutually exclusive, and phrased the way your buyers would describe them.
- Keep attributes independent. A level in one attribute should not imply a level in another, or the model cannot separate their effects.
- Always include price if pricing is part of the decision, with levels that bracket what you believe the market will bear.
Here is a simple example for a B2B analytics product:
| Attribute | Level 1 | Level 2 | Level 3 |
|---|---|---|---|
| Price per seat / month | $25 | $50 | $75 |
| Integrations | 10 apps | 50 apps | 200+ apps |
| Support | Email only | Email + chat | Dedicated manager |
| Data refresh | Daily | Hourly | Real time |
Every choice task the respondent sees will be built by combining one level from each row into competing product profiles. Before you finalize the list, sanity-check the attributes and levels with a few real buyers or a quick round of qualitative interviews, so you are not testing options nobody cares about. The guide on how to turn product research into better product decisions is a useful reference for framing which trade-offs are actually worth measuring.
Step 2: Design the choice tasks
A choice task is a single screen where the respondent picks between two to four product profiles. A typical CBC study asks each respondent to complete 8 to 15 of these tasks.
You almost never show every possible combination. With the four attributes above, there are 3 x 3 x 3 x 3 = 81 possible profiles, and larger designs run into the thousands. Instead, conjoint software generates an experimental design, a carefully balanced subset of profiles that still lets the model estimate every part-worth utility. Good designs aim for two properties: level balance, so each level appears roughly equally often, and orthogonality, so attributes vary independently of one another.
Practical guidelines for the choice tasks:
- Show three to four profiles per screen for most B2B and B2C products.
- Include a “none of these” option so you can detect when the whole set is unappealing, which matters a great deal for pricing.
- Randomize the order of tasks and profiles across respondents to avoid position bias.
- Keep the profile descriptions short and scannable so people read them rather than skim.
Modern conjoint tools handle the experimental design automatically, but you should still review a few sample tasks yourself to confirm the profiles read like real products a buyer might compare.
Step 3: Size your sample
Conjoint estimates are only as stable as the data behind them. A common working minimum is 200 to 300 completed responses for a single market. If you plan to report results separately by segment, such as small business versus enterprise buyers, size each segment to that minimum on its own rather than splitting one small sample.
Sample size scales with design complexity. More attributes and levels mean more parameters to estimate, which requires either more respondents or more choice tasks per respondent. As a rough guide:
| Study scope | Attributes | Suggested completes |
|---|---|---|
| Simple, single market | 3 to 4 | 200 to 300 |
| Standard, single market | 4 to 6 | 300 to 400 |
| Multiple segments reported separately | 4 to 6 | 300+ per segment |
Just as important as the count is who those respondents are. Conjoint asks people to make realistic buying trade-offs, so the sample needs to be people who actually buy, evaluate, or approve purchases in the category. General panelists with no purchase context will still click through the choices, but the utilities they produce will not reflect how the real market behaves. This is where sample quality quietly makes or breaks the study, and it is worth reading up on how to recruit B2B research participants if your buyers are professionals rather than general consumers.
Step 4: Read the part-worth utilities and importance scores
Once responses are in, conjoint software estimates two headline outputs.
Part-worth utilities are scores for each level of each attribute. A higher utility means the level is more preferred. Utilities are relative, so what matters is the difference between levels, not the absolute number. For price, you will typically see utility fall as price rises, and the shape of that decline tells you how sensitive buyers are.
Attribute importance is derived from the utilities. For each attribute, you take the range between its highest and lowest level utility, then express each attribute’s range as a percentage of the total. An attribute with a wide utility range drives choices strongly; a narrow range means buyers barely care. A simplified read might look like this:
| Attribute | Utility range | Importance |
|---|---|---|
| Price | Wide | 40% |
| Integrations | Moderate | 28% |
| Data refresh | Moderate | 20% |
| Support | Narrow | 12% |
In this example, price and integrations dominate the decision, while support level barely moves buyers. That single table can redirect a roadmap: investing heavily in premium support would not shift demand much, whereas expanding integrations would. The canonical reference for how these models are estimated is well documented in the Sawtooth Software conjoint literature and summarized in the general overview of conjoint analysis on Wikipedia.
Step 5: Use the results for pricing and packaging
The real payoff of conjoint is the market simulator. Because you have a utility for every level, you can define any product configuration, plus competitor configurations, and estimate the share of buyers who would choose each one. This lets you answer concrete business questions.
- Pricing. Because price is an attribute with its own utilities, you can model how demand shifts as you move up or down the price ladder, and find the point that balances share and revenue. Conjoint complements dedicated pricing methods such as the Van Westendorp price sensitivity survey, which is faster to field when price is the only question you have.
- Feature prioritization. Importance scores tell you which features actually change buying decisions, so you can defer low-utility work. Pair this with a Kano model feature prioritization analysis to separate features that delight from features that are merely expected.
- Packaging and tiers. Simulate different bundles to design good-better-best tiers that each capture a distinct slice of demand without cannibalizing one another.
- Segmentation. Run the simulator within segments to see where willingness to pay and feature priorities diverge, which feeds directly into a customer segmentation study.
Every one of these uses assumes the sample reflects the real buying population, which is why recruitment is not a footnote to conjoint but a precondition for trusting it.
Conjoint analysis versus MaxDiff
Conjoint and MaxDiff are often confused because both are trade-off methods, but they answer different questions.
| Conjoint analysis | MaxDiff | |
|---|---|---|
| What respondents do | Choose between full product profiles | Pick most and least important from short lists |
| Handles price | Yes, price is an attribute | No, not natively |
| Main output | Part-worth utilities and market simulations | A clean importance ranking of items |
| Best for | Pricing, packaging, configuration decisions | Prioritizing a long list of features or messages |
| Complexity | Higher | Lower |
The short version: use conjoint when price and product configuration are central to the decision and you want to simulate market outcomes. Use MaxDiff when you simply need to rank a long list of features, benefits, or messages and price is not part of the question. Many teams run MaxDiff first to shortlist features, then conjoint to price and package the winners. Both fit within the broader toolkit covered in the basics of market research.
Recruiting the right participants
The most common reason a conjoint study misleads a team is not a modeling error, it is the sample. If the people answering are not genuine buyers of the category, the trade-offs they make are hypothetical, and the pricing and packaging decisions built on top inherit that flaw.
This is the hard part of conjoint, especially in B2B, where the population you need might be a few thousand people with a specific role, budget authority, and recent purchase experience. General consumer panels rarely reach them at the quality you need.
CleverX is a B2B research platform built for exactly this problem, with a network of more than 8 million verified professionals plus B2C reach across 150 or more countries. Every participant is identity and employment verified, so a conjoint study aimed at, say, IT security buyers actually reaches IT security buyers rather than people who selected that industry on a signup form. Studies are pay-as-you-go with roughly 2 to 5 day delivery, which means you can field a representative conjoint sample in the same week you finalize the design. When your entire pricing or packaging decision rests on how real buyers trade off attributes, that sample quality is what makes the utilities defensible in a board meeting.
Ready to run a conjoint study on a representative sample? Recruit verified participants on CleverX and get trade-off data from the buyers who actually make the decision.
Frequently asked questions
What is conjoint analysis used for?
Conjoint analysis measures how buyers make trade-offs between product attributes such as price, features, and brand. By asking people to choose between realistic product profiles, it estimates how much each attribute and level contributes to preference. Teams use it to set pricing, decide which features to build, design packages and tiers, and simulate how the market would respond to a new configuration before it launches.
What is the difference between conjoint analysis and MaxDiff?
Conjoint asks respondents to choose between full product profiles that combine several attributes at once, so it captures trade-offs including price and outputs part-worth utilities you can use for market simulations. MaxDiff asks respondents to pick the most and least important items from short lists of single features, which is simpler and better for ranking a long list of features or messages. Use conjoint when price and configuration matter, and MaxDiff when you just need a clean priority order.
How many attributes and levels should a conjoint study have?
Most well-designed choice-based conjoint studies use four to six attributes, with two to five levels each. Beyond six attributes respondents start to simplify their decisions, which adds noise. Keep attributes independent, make every level realistic and mutually exclusive, and describe them in plain language your buyers would actually use.
How many respondents do you need for conjoint analysis?
A common working minimum is 200 to 300 completed responses for a single market, with 300 or more preferred when you plan to analyze results by segment. Larger designs with many attributes and levels need more respondents to estimate stable utilities. If you plan to report separate results for two or three buyer segments, size each segment to that minimum on its own.
What are part-worth utilities in conjoint analysis?
Part-worth utilities are the scores conjoint estimates for each level of each attribute, showing how much that level adds to or subtracts from a buyer’s preference. Higher utility means a more preferred level. Comparing the range of utilities within an attribute tells you its importance, and comparing levels within an attribute tells you the relative appeal of each option, including how much value buyers assign to a lower price.
Can conjoint analysis be used for B2B pricing?
Yes. Conjoint is widely used for B2B pricing, packaging, and feature decisions, but it depends on reaching real buyers who understand the trade-offs, such as economic buyers and strong influencers with recent purchase context. The main challenge in B2B is recruiting a representative sample of qualified professionals rather than general panelists, which is why buyer verification matters more than raw panel size.