Adaptive conjoint analysis (ACA) is a conjoint method that tailors each survey to the person taking it. It learns which product features matter most to a respondent, then spends the remaining questions on those features. That makes it practical for studies with many attributes.
ACA is also less common than it once was. Choice-based methods now dominate pricing work, and ACA has known limits. Knowing where it still fits saves time and budget.
In this guide, we will explain how ACA works, how it differs from CBC and ACBC, and when to use it.
What is adaptive conjoint analysis?
Adaptive conjoint analysis (ACA) is a survey method that customizes each respondent’s interview, so the questions focus on the product features that person values most. The word “adaptive” means the software re-estimates preferences after each answer and picks the next question to learn as much as possible.
ACA belongs to the wider family of conjoint analysis, which measures how people trade off product features. Three terms come up throughout this guide:
- Attribute: A product feature being tested, such as price or brand
- Level: One option within an attribute, such as $20 or $30 per month
- Part-worth utility: A score that shows how much a level adds to, or subtracts from, a person’s preference
ACA produces a part-worth utility for every level, for every respondent.
Sawtooth Software released ACA in 1985. Its own site now reports that ACA makes up only about 2% of conjoint projects among its users. ACA can run on the web or on offline devices in CAPI mode, short for computer-assisted personal interviewing.
How does adaptive conjoint analysis work?
ACA runs in two main parts. A short self-explicated section comes first, followed by adaptive paired comparisons. Self-explicated means respondents state their preferences directly instead of choosing between products.
- Rank the levels.
For each attribute, the respondent orders the levels from most to least preferred. - Rate importance.
The respondent says how much the gap between the best and worst level matters. These answers form the starting estimate, called priors. - Compare partial concepts in pairs.
The software shows two concepts, each described on only some attributes. That is a partial profile, and it can use up to five attributes per concept. The respondent rates which concept they prefer on a graded scale. - Update after each page of pairs.
The system refines the estimate, so later pairs probe the trade-offs that matter most to that person. - Check with calibration concepts (optional).
The software builds a few concepts from the most important attributes and asks the respondent to rate each one. - Estimate the part-worths.
The software combines priors and pairs into one set of utilities per respondent. Many studies use hierarchical Bayes (HB) estimation, a statistical method that uses the whole sample to steady each person’s estimates.
Adaptive conjoint analysis example
Consider a software company testing ten features for a new project management tool. The figures below are illustrative, not from a real study. One respondent’s stated importance ratings (1 to 4) might look like this:
- Automation: 4, shown often in pairs
- Integrations: 4, shown often in pairs
- Price: 3, shown often in pairs
- Reporting: 3, shown sometimes
- Mobile app and storage limit: 1, shown rarely
This respondent cares about automation, integrations, and price, so most pairs pit those against each other. A pair might show Plan A with advanced automation at $30 against Plan B with basic automation at $20. The respondent rates which plan they prefer and by how much.
A standard choice-based study would repeat all ten attributes in every task. Here, most of the interview goes to the four attributes that matter to this person. Across hundreds of respondents, the software averages the utilities and might report that automation drives 28% of preference while price drives 18%.
Adaptive conjoint analysis vs choice-based conjoint
Choice-based conjoint (CBC) asks respondents to pick their favorite from sets of complete product profiles. Adaptive choice-based conjoint (ACBC) adds adaptive stages to a choice design. MaxDiff, short for maximum difference scaling, asks people to pick the best and worst item from small sets of items.
| Method | What respondents do | Best fit | Weak spot |
|---|---|---|---|
| ACA | Rate importance, then compare partial concepts in graded pairs | Many attributes, product design, segmentation | Understates price effects, main effects only |
| CBC | Pick one full product from each set | About six or fewer attributes, pricing | Fatigue as attributes grow |
| ACBC | Build their own product, screen concepts, then choose in a tournament | About five or more attributes, deeper individual insight | Longer surveys |
| MaxDiff | Pick the best and worst item from a set | Ranking 12 to 100+ items | Does not test bundles or price |
ACA is a ratings-based approach with partial profiles, while CBC asks for a choice among full profiles. It favors CBC when pricing research is a big part of a study. An ACBC survey is often two to three times as long as a traditional CBC survey.
For a deeper look at the MaxDiff question, see this comparison of MaxDiff vs conjoint.
Types of adaptive conjoint analysis
Three approaches carry the adaptive label in practice. Each adapts the interview in a different way. For the wider map of methods, see the guide to types of conjoint analysis.
Classic ACA
Classic ACA combines self-explicated priors with graded paired comparisons. It was built for studies with more attributes than older methods could handle. Its key traits are:
- Up to 30 attributes and 15 levels per attribute in the larger system
- Interviews that focus on a few attributes at a time
- Individual-level utilities for every respondent
ACA with hierarchical Bayes (ACA/HB)
ACA/HB keeps the interview and changes the estimation. Hierarchical Bayes uses information from the whole sample to improve each individual’s utilities. The ACA/HB technical paper notes that the module does not use the calibration concepts section.
Adaptive choice-based conjoint (ACBC)
ACBC blends adaptivity with choice tasks. It has three core stages, and it recommends the method for about five or more attributes.
- Build your own: The respondent configures their ideal product.
- Screening: The respondent reviews concepts near that ideal and marks which are possibilities.
- Choice tournament: Winning concepts advance and losing concepts drop out. Each task shows at most three concepts.
Pros and cons of adaptive conjoint analysis
ACA’s strengths come from its adaptivity. Its weaknesses come from its age and design. Both matter when you pick a method.
Pros
- Handles long attribute lists, up to 30 in the larger ACA system
- Keeps interviews focused on the attributes each person rates as important, which limits information overload
- Produces individual-level utilities that support segmentation
- Includes a simulator for what-if tests on price, product formulation, or marketing changes
- Runs on the web or offline in CAPI mode
Cons
- It is a weak methodology for price studies
- Tends to understate the importance of price
- Assumes a main-effects model, so attributes do not interact
- Relies on a self-explicated importance question that researchers have called a weak link
- Shows partial concepts, so tasks mirror shopping less closely than CBC choices
When to use adaptive conjoint analysis
Use ACA when the attribute list is long and price is not the headline question. The table below maps common situations to a better-fitting method.
| Your situation | Better fit | Why |
|---|---|---|
| Price is the main question | CBC or ACBC | ACA is weak for price studies |
| Six or fewer attributes | CBC | Full profiles mirror shopping |
| 5 to 12 attributes, deeper insight needed | ACBC | Adaptive stages probe each person’s decision rules |
| More than 12 attributes, price is secondary | ACA, or a MaxDiff screen first | ACA supports up to 30 attributes |
| In-person or offline interviews | ACA in CAPI mode | Runs without a web connection |
| Small sample | ACBC over CBC | Performs better with small samples |
Partial-profile CBC is a non-adaptive alternative. It shows only some attributes per task and works when attribute lists are long. If you are still shortlisting software, compare conjoint analysis platforms and the designs each one supports.
Real-world uses of adaptive conjoint analysis
ACA fits decisions with many moving parts. Healthcare, technology, and consumer products are the most common settings.
Healthcare and patient preferences
A peer-reviewed study tested ACA as a way to assess patient preferences in oncology. The authors describe it as an individually tailored technique that asks patients to trade off treatment dimensions. Typical attributes include:
- Side effects
- Success rates
- Cost
- Convenience of care
In the US, the FDA finalized guidance on voluntary patient preference information for medical device review in 2016.
B2B and technology products
Complex products carry many features and long buying processes. An academic comparison found that ACBC suits more complex products or services and gives deeper insight, such as second-best options and unacceptable features. Adaptive methods help here because no buyer cares about every feature.
Consumer product design
Consumer brands use adaptive designs to cut long feature lists down to a shortlist. Common goals include:
- Picking which features to build first
- Finding segments with different priorities
- Sizing the premium a brand can charge
How to measure and evaluate adaptive conjoint results
Good ACA results are stable, logical, and able to predict real choices. Six checks cover the basics.
| Check | What it tells you | Criterion |
|---|---|---|
| Relative importance | Share of the decision each attribute drives | Shares total 100%, and price matches business expectations |
| Part-worth order | Value of each level | Levels follow a logical order, such as lower price preferred |
| Holdout fit | Whether the model predicts real choices | Use calibration concepts as holdouts |
| Sample size | Stability of estimates | About 200 to 300 completes for one group, more per segment |
| Design limits | Whether the study fits the method | Up to 30 attributes, 15 levels each, five attributes per pair |
| Drop-off | Respondent fatigue | Completion at the pairs section matches earlier sections |
The ACA/HB paper says calibration concepts can serve as holdout concepts for testing prediction accuracy. For sample size, QuestionPro’s conjoint guide suggests 200 to 300 completes for a single-group study.
Market simulation
A market simulation turns utilities into predictions. The researcher describes a set of products by their levels on each attribute. The software uses respondent utilities to estimate preference shares or average buying likelihoods for each product, then averages the results across respondents.
Common mistakes in adaptive conjoint analysis
Most ACA problems trace back to method choice and design shortcuts. This table lists the failures that appear most often.
| Mistake | Why it hurts | What to do |
|---|---|---|
| Using ACA for pricing decisions | It tends to understate price importance | Use CBC or ACBC when price drives the result |
| Trusting stated importance alone | Self-explicated ratings can flatten differences | Review priors and pairs together |
| Assuming attributes interact | ACA is a main-effects model | Combine two attributes into one when interaction matters |
| Skipping validation | You cannot judge prediction accuracy | Use calibration concepts or holdout tasks |
| Overloading each pair | Concepts hold five attributes at most | Keep pairs short and attributes distinct |
| Ignoring segment sample sizes | Group results become unstable | Plan completes for your smallest segment |
How QuestionPro supports conjoint studies
QuestionPro’s conjoint analysis tool is built around choice-based designs. Researchers enter attributes and levels, and the platform builds the tasks respondents see.
Design types and reports
Three design types are available:
- Random: Builds a unique combination of attribute levels for each respondent
- D-optimal: Uses an algorithm to estimate parameters with minimum variance for your sample size
- Import: Brings in SPSS-format designs, such as fractional factorial orthogonal designs
Reports cover attribute importance, profiles, estimated brand premium, and price elasticity. The market segmentation simulator forecasts the share of products that do not exist yet. You can also filter results by location, demographics, and device type.
Studies with many attributes
For long attribute lists, a screening path works well inside QuestionPro’s Market Research Software:
- Run a MaxDiff study to rank the full list of candidate attributes.
- Build a choice-based conjoint study around the top attributes.
- Simulate scenarios to test pricing and product changes.
Choosing how adaptive your study needs to be
Adaptive conjoint analysis earns its place when a study has many attributes and price is not the main question. In those cases, it keeps interviews focused and still delivers individual-level utilities. When pricing drives the decision, choice-based methods remain the safer foundation.
Start by counting your attributes, then decide how much adaptivity your questions and your budget can support.
Frequently Asked Questions (FAQs)
No. In market research, ACA stands for adaptive conjoint analysis. The Affordable Care Act is a US health law with the same initials. In healthcare reports, spell out each term on first use so readers never confuse the two.
ACBC does better with small samples than CBC. A five-attribute CBC study is feasible with about 100 respondents. Segment-level reads still need enough completes in each group.
Plan for a longer interview than a standard choice study. An ACBC survey is often two to three times as long as a traditional CBC survey, so budget respondent incentives and drop-off accordingly.
It can include price as an attribute, but results need care. ACA often understates price importance. CBC or ACBC are safer for dollar estimates.
Lead with relative importance, then show part-worths for the top two attributes. Add one market simulation that compares today’s product with the proposed change. Keep full utility tables in an appendix so executives see the decision, not the model.
