Choice-based conjoint (CBC) and adaptive choice-based conjoint (ACBC) both measure how much people value each feature of a product by watching them choose. They differ in how they get there. CBC shows every respondent a fixed set of choice tasks. ACBC builds each interview around the person taking it. This guide explains how each works, what the published evidence says, and how to decide between them.
What CBC is
Choice tasks
In a CBC study each respondent sees a series of screens, each showing a few product concepts built from your attributes (brand, size, features, price). They pick the one they would buy, or none. Across a dozen or so screens the design varies every attribute independently, so a model can work out how much each level pushes choice up or down.
Why it is the default
CBC is quick to answer, easy to explain to stakeholders and close to how people shop: a shelf or a comparison page shows a few options and you choose one. With a modest number of attributes it produces reliable part-worths for each respondent when estimated with Hierarchical Bayes.
What ACBC is
ACBC was developed by Sawtooth Software for studies where a fixed choice grid asks too much of people. Its technical paper notes that CBC respondents typically spend about 12–15 seconds per task, and that in one nine-attribute study 85% of choices could be explained by attention to no more than four attribute levels (Sawtooth Software, 2014). When products get complex, people simplify. ACBC is designed to capture how they simplify.
Build-your-own
The interview starts by asking respondents to configure the product they would most like, trading features against price. This anchors the rest of the interview around what is relevant to them.
Screening, unacceptables and must-haves
Next, respondents see a few concepts per screen near their preferred configuration and mark each as "a possibility" or not. Sawtooth's design uses about seven of these screens with three to five concepts each. Along the way, the software asks whether certain levels are unacceptable or must-haves, which mirrors the non-compensatory rules people use to cut a market down to a shortlist (Sawtooth Software, 2014).
The choice tournament
The concepts a respondent kept compete in a tournament of choice tasks, typically shown three at a time, until a winner emerges. Because every concept in the tournament already passed that person's screen, the trade-offs are between products they would genuinely consider.
When each is better
Number of attributes
Sawtooth's own guidance is that ACBC is "best applied" to studies with about five attributes or more, and that small studies such as brand, pack and price "wouldn't seem to benefit". It recommends staying within about 12 attributes of up to about seven levels each (Sawtooth Software, 2014).
Respondent time and engagement
ACBC interviews are longer. In the comparisons reported in the technical paper they took 50–200% longer than CBC, but respondents rated them as more interesting and more realistic (Sawtooth Software, 2014).
Sample size and accuracy
On custom holdout tasks, ACBC predicted choices better than CBC in the laptop study (61% vs 50% hit rate) and the housing study (44% vs 37%), and tied with it in a four-attribute drive-through study (Sawtooth Software, 2014). An independent comparison by Chapman et al. (2009) found broadly similar estimates from the two methods, with ACBC showing greater price sensitivity that agreed closely with real market data. Orme (2019) adds that when the sample is small and the attribute list is long, adaptive methods may be the better tool.
Where CBC still wins
Few attributes. With four or five attributes there is little for ACBC's screening to simplify, and the evidence above shows no accuracy gain.
Shorter interviews. If survey length or cost per complete is the binding constraint, CBC gets more people through.
Portability. A CBC design is a fixed set of tasks, so it can be fielded and checked in many tools. ACBC needs software that generates each interview on the fly.
How Ordeen decides
The seven-attribute rule
Ordeen recommends the method when it drafts your design. A CBC study with more than seven attributes is routed to ACBC; seven or fewer stays CBC. Only attributes with at least two levels count, because an attribute with one level carries no trade-off. The Design tab explains the recommendation, and you can review it before launch. MaxDiff and menu-based choice (MBC) are chosen when you create the project and are never switched automatically.
The cut-off sits above Sawtooth's "about five attributes" because ACBC's longer interview is a real cost. Seven is Ordeen's product policy, not an industry standard.
What counts as an attribute
An attribute is any product dimension that varies between concepts: brand, size, a feature, a service level, price. Splitting one idea across two attributes (for example "has feature X" and "feature X quality") inflates the count and can push a study into ACBC unnecessarily, so it is worth merging those before you review the design.
How Ordeen estimates ACBC
Each respondent's own tournament
Because every ACBC respondent sees different concepts and a different number of tournament tasks, Ordeen rebuilds each person's design from the interview itself and estimates on exactly the alternatives they were shown. Tasks a respondent never saw are excluded from the likelihood rather than filled in.
Screener likelihood
The screening answers enter the model too. Each "possibility" or "not for me" judgement is modelled as the concept's utility compared with a personal cutoff, and those cutoffs are estimated hierarchically alongside the part-worths, with Hierarchical Bayes.
None threshold
Each respondent also gets their own utility for choosing none, so simulations can show how many people would walk away rather than buy any product on offer.
One difference from Sawtooth's implementation: Sawtooth also codes the build-your-own answers as choice tasks (Sawtooth Software, 2014). Ordeen uses the build-your-own section to centre the interview, and estimates from the screening and tournament answers.
Can ACBC run in another survey tool?
Not easily. Each ACBC interview is generated live from the respondent's earlier answers, so it cannot be exported as a static questionnaire. In Ordeen, ACBC studies run on Ordeen's own survey, with the integrated panel or with a link you send to your own audience.
Summary checklist
Four or five attributes, simple products, tight interview budget: CBC.
Eight or more attributes, considered purchases, or customers who shortlist before they compare: ACBC.
Either way, keep each attribute to a single idea and check the design report before launch.
References
Sawtooth Software (2014). ACBC Technical Paper. Sawtooth Software Technical Paper Series. sawtoothsoftware.com
Chapman, C. N., Alford, J. L., Johnson, C., Weidemann, R. and Lahav, M. (2009). CBC vs. ACBC: Comparing Results with Real Product Selection. Proceedings of the Sawtooth Software Conference. PDF
Orme, B. (2019). Getting Started with Conjoint Analysis, 4th ed. Research Publishers LLC.
