When you ask customers directly what they expect from a product, most will say they want “high quality at a low price.” The problem is that survey declarations rarely align with actual purchasing decisions, yet it is those decisions that determine whether an implementation succeeds. Conjoint analysis resolves this disconnect by requiring respondents to make trade-offs between product features – exactly as they do in a store or during a B2B purchasing process.
What is conjoint analysis and when is it worth using?
Conjoint analysis is a family of measurement methods based on the assumption that product preferences can be decomposed into part-worth utilities assigned to individual attributes and their levels. Rather than evaluating features in isolation, respondents choose between complete product profiles that differ across several parameters at once. The same logic applies to real-life decision-making: consumers do not choose “price” or “brand” separately, but rather the offer as a whole.
Conjoint analysis is particularly effective when the goal is to design a new product, reposition an existing portfolio, set pricing packages, optimize variant configurations, or estimate willingness to pay for individual features. In B2B projects, conjoint helps distinguish the impact of factors such as contract length, SLA level, billing model, or scope of technical support. In B2C, it most often concerns consumer product features, packaging, distribution channels, or the structure of a subscription offer.
It is worth distinguishing between the main variants of the method. Choice-Based Conjoint (CBC) presents respondents with sets of alternatives from which they select one option, or none – making it the closest to actual consumer behavior. Adaptive Choice-Based Conjoint (ACBC) adjusts subsequent tasks based on earlier responses, allowing more attributes to be studied without causing cognitive overload. MaxDiff (best-worst scaling), while technically not a classic conjoint method, is often used to prioritize a list of features when there is no need to model complete profiles. Menu-Based Conjoint reflects situations in which customers assemble an offer themselves from modules.
How to design a conjoint study step by step?
The quality of conjoint analysis results depends first and foremost on how product attributes and their levels are defined. This is a workshop stage involving both the research team and business representatives familiar with the realities of the category. Attributes must be distinct, understandable to respondents, and genuinely influence purchasing decisions – including cosmetic features dilutes the model.
A typical process for designing a conjoint customer preference study involves several steps that should be completed in sequence:
- identifying attributes based on desk research, exploratory interviews, and an analysis of competitors’ offers,
- defining the levels of each attribute – realistic, feasible to implement, and covering the market range,
- selecting a method variant (CBC, ACBC, MaxDiff, MBC) suited to the number of attributes and the nature of the decision,
- generating an experimental design that ensures the best possible statistical efficiency, balance, and minimal multicollinearity,
- conducting a pilot study with a small sample to verify task comprehension and completion time,
- conducting fieldwork with a sample selected from the defined population of decision-makers,
- estimating the model, most often using Hierarchical Bayes, and running market simulations.
The number of attributes is a frequent point of contention. In classic CBC, researchers typically work with six to eight attributes; beyond that number, the risk increases that respondents will begin using simplified heuristics, for example, looking only at price. ACBC makes it possible to expand this range safely, but at the cost of a longer interview. The choice of method should result from a map of actual decision criteria in the category, not from a desire to “study everything at once.”
As Hume’s Institute experts point out, customers rarely choose a product for the reasons they state in a declarative survey – conjoint requires them to make real choices between alternatives, and only then does the model reconstruct the hierarchy of feature importance from those choices. This is the fundamental difference from asking “how important is price to you on a scale of 1-5?”, to which almost everyone will answer “very important,” adding nothing to the product decision.
The result of estimation is a set of part-worth utility values for each level of every attribute, usually estimated at the individual level. This makes it possible to do three things: calculate the relative importance of attributes, estimate willingness to pay for a specific feature by converting utility into monetary units using price levels, and run a market simulator that forecasts preference shares for any configurations of products competing in the market. In Hume’s Institute projects, it is observed that the simulator, rather than the report itself, is most often the research output used – because it allows product and pricing scenarios to be tested without conducting additional studies.
What are the most common mistakes and limitations of conjoint analysis?
The most serious mistake is treating conjoint as a universal diagnostic tool. The method measures preferences under conditions of forced choice between predefined profiles – it will not identify an attribute that the researcher did not include in the design. If a decisive purchasing factor, such as a specific technical feature or availability through a local channel, is not on the list, the model simply will not see it. That is why conjoint should always be accompanied by an exploratory phase – qualitative interviews, desk research, and review analysis – to map the customer’s decision-making space.
The second mistake is using unrealistic attribute levels. If the price range in the design is too narrow, the model will consider price relatively unimportant. If it is too wide, the model will artificially overstate its importance. Levels must reflect the actual range of market offers and what the company can realistically implement. The same applies to qualitative levels: descriptions must be specific enough for respondents to understand them consistently, but general enough not to suggest a particular brand.
The third pitfall concerns the sample. Conjoint requires a sample large enough to estimate utilities reliably at the individual level – for CBC, this usually means at least several hundred respondents per segment to be analyzed separately. A sample that is too small will not allow preference segments to be identified reliably or simulations to be run for niche groups. Respondent selection also matters: a customer preference study only makes sense when participants genuinely belong to the population of decision-makers in the category.
A limitation of the method is also that it measures stated preferences in the context of a choice task, rather than actual purchasing behavior. The retail context, time pressure, shelf availability, and promotional activities – these elements are not present in classic conjoint. This is why good projects combine conjoint with other methods: in-store tests, transaction data analysis, and pricing experiments. Conjoint analysis answers the question, “what value does a respondent assign to a product feature under conditions of deliberate comparison?”, rather than “what will they buy at 5:00 p.m. on Friday on their way home from work?”
It is also worth mentioning the comparison with simpler methods. A declarative survey asking about the importance of attributes is less expensive and faster, but produces the well-known flattening effect – respondents rate most features as important. Attribute ranking provides a hierarchy, but does not show how much more important one feature is than another, nor does it allow market simulations. MaxDiff addresses the flattening problem, but does not make it possible to study interactions between attribute levels in complete profiles, such as price and brand. Conjoint remains one of the few survey methods that simultaneously quantifies feature importance, assigns a monetary value to individual levels, and enables preference share forecasting.
When does conjoint deliver the most value? Typical applications
Below is a summary of research situations in which conjoint typically provides results that are difficult to obtain through other methods:
- designing a new product or service before market launch, when it is necessary to select a feature configuration that maximizes interest among a defined target group,
- setting the pricing structure and packages, such as basic/standard/premium versions, in subscription and SaaS models,
- estimating willingness to pay for individual features – crucial when deciding which ones to develop and which to discontinue,
- preference segmentation – identifying customer groups that differ in their hierarchy of attribute importance,
- competitive analysis and preference share simulation when changing one’s own offer or competitors’ offers,
- optimizing B2B contracts, where commercial, technical, and service terms must be selected simultaneously.
Frequently asked questions
How does conjoint differ from a standard preference survey?
A preference survey asks directly about the importance of features, most often using rating scales, which leads to a flattening effect – most attributes appear important. Conjoint does not ask about importance. Instead, it observes choices between complete product profiles and reconstructs the hierarchy of preferences from them. As a result, findings better reflect actual trade-offs than declarations and make it possible to build a preference share simulator.
Which product attributes can be studied using conjoint?
Virtually any feature that can be described clearly and presented to respondents as several distinct levels – price, brand, technical parameters, package size, contract length, support level, delivery channel, or design. The limitations are the number of attributes, usually 6-8 in classic CBC and more in ACBC, and the need for each attribute to be understandable to the target group. Emotional and image-related features can also be studied if they are translated into specific, comparable descriptions.
When does conjoint replace a focus group?
Never completely – these methods are complementary rather than competing. A focus group serves an exploratory purpose: it reveals which attributes matter at all, the language customers use to discuss them, and the perceptual barriers that emerge. Conjoint comes later and quantifies what has been identified qualitatively. Skipping the qualitative phase increases the risk that the conjoint design will omit an attribute crucial to the purchasing decision.
Ask about a preference study using conjoint analysis
If you are considering measuring customers’ actual preferences regarding product features, an offer, or a pricing structure, the Hume’s Institute team can help select a conjoint variant suited to the specific business decision and design the study from attribute workshops through to the market simulator. Contact us to discuss the project scope and delivery timeline.