How Conjoint Shows You Which Features Drive Choice and What People Will Pay
By aytm
- article
- Advertising Testing
- Agile Quantitative Research
- Brand Tracking
- Conversational AI
- DIY Surveys
- Insight Communities
- Pricing Research
- Product Development Research
- Survey Research
- Survey Software
- Concept Testing
- Conjoint Analysis
Summarise with AI
This is the third article in aytm’s Innovation Intelligence series, which explores each stage of the innovation research funnel. Each piece draws on the Product Innovation curriculum in the aytm Lighthouse Academy.
Explore the full suite of innovation courses, including the Choice-Based Conjoint Advanced Certification, here.
You can know exactly which features drive the choice to buy, which attributes justify charging more and how those parts combine into the product people actually pick. That’s the question every product team is really asking, and choice-based conjoint analysis was built to answer it. It reveals your preference architecture: which features earn the sale, which are table stakes that win no credit but cost you the deal when they’re missing and which attributes carry a price premium. Those are questions about trade-offs, and trade-offs are what conjoint is designed to surface.
Ask consumers what they want in a product, and you’ll hear the same answer every time: everything. More features, higher quality and lower price. Put any one of those on a rating scale, and it scores well, because there’s no cost to saying yes. Conjoint adds the cost back in, so the data finally tells you what to build.
Why Conjoint Sees What Rating Scales Can’t
Conjoint reveals what people will trade, and that’s the whole game. When you ask a respondent to rate features one at a time on a polarity scale or a progressive matrix, each feature gets evaluated in isolation. The respondent never has to give anything up to get something else, so the data stays quiet about real preference. As aytm’s Choice-Based Conjoint curriculum puts it directly, pasting together individually rated features “may not result in a truly optimal product, because each feature would be evaluated in isolation”.
Conjoint inverts the task. Respondents see full product profiles, two to five alternatives at a time, typically with a “None” option to walk away, and they pick the one they’d most likely buy. That single choice forces every implicit comparison at once. Is the better screen worth the higher price? Is free shipping worth more than a faster delivery window? Across a series of these tasks, the pattern of choices lets the model recover the value, or importance, of each component part in terms of its actual contribution to the decision. The technique distils a product into its constituent parts, tests them in combination and recovers which parts customers prefer and by what magnitude.
That’s the point. Conjoint, in the words of its discipline, is a way “to determine the hidden rules consumers use to make trade-offs among different products and to quantify the values they place on different features.” Rating scales ask people what they want, but conjoint watches what they choose.
The Vocabulary of Attributes, Levels, and Utilities
Attributes are the features that make up the product. For a vacation package, that might be price per night, location, Wi-Fi, breakfast and number of nights. Levels are the options within each attribute so that location could be beach, mountains, city or country. You build the list of attributes first, then the levels inside each one, and the platform constructs the choice tasks from there.
Conjoint analysis was adapted to market research by Paul Green at the Wharton School in the early 1970s, and it matured over the decades. Jordan Louviere developed the choice-task approach, Richard Johnson contributed pairwise trade-off analysis, Valerii Federov created the exchange algorithm for efficient designs and Greg Allenby’s work brought the Hierarchical Bayesian techniques that let the model estimate preferences for each individual respondent rather than only the crowd. That last advance is what made conjoint usable for segmentation. The history matters because it explains why you can trust the output: this is a fifty-year-old technique with deep statistical foundations.
How Aytm Runs It In Express and Segmentation
aytm offers choice-based conjoint in two analysis modes, and choosing between them is the first real decision you make. Both build the experimental design the same way, using Federov’s exchange algorithm to produce a design that’s D-optimal and nearly orthogonal, presenting each level an approximately equal number of times, often across more than a hundred versions for larger setups.
Express mode uses a Multinomial Logit model and analyses all respondents in a single pooled sample. It returns directional insight on the aggregate, with a shorter respondent task and a faster, cheaper read. Segmentation mode uses a Hierarchical Bayesian Multinomial Logit model that estimates utilities at the individual-respondent level, then runs a two-stage process. A Gaussian Mixture Clustering model groups respondents with similar preferences, and a second pass identifies which traits and survey answers explain the differences between clusters. The output includes auto-generated Personas.
The published guide frames the choice well: it’s “like choosing between a snapshot and a detailed portrait”. Segmentation is the platform’s default recommendation for anything beyond a low-priority question, and the cheese problem from the curriculum shows why. If men dislike a topping and women like it, Express records the aggregate, lands somewhere in the middle and leaves you guessing. Segmentation discovers the two distinct groups and lets you act on each. The trade-off is worth noting: Segmentation asks more tasks per respondent and more sample. aytm recommends roughly 750 to 1,500 completes for a robust conjoint study.
Reading the Output
A fielded conjoint produces four to five interpretable outputs, and knowing what each one does is the difference between a finding and a slide nobody trusts.
- Utility scores are zero-centred part-worth values showing the relative value respondents place on each level. Sum the part-worths of a full profile to get its total utility.
- Attribute importance shows which attribute moves the decision most. In the curriculum’s vacation-package study, price per night came out first, followed by location, the kind of ranking a rating scale tends to bury because everything rates highly.
- Preference likelihood expresses the relative preference of a given combination against all others tested, on a 0-100 scale. Read it relatively: a product at 75 is 1.5 times as likely to be chosen as one at 50. It expresses preference, so treat it as a preference rather than a market forecast.
- Market simulator is the output to lead with. It lets you build up to seven hypothetical products, pit them head-to-head (or against the “None” walk-away option), and watch preference share shift as you change a level. Because it uses preference share, it’s the closest thing to simulating a real shelf decision, and it exports to an interactive Excel file.
One honest limit, stated plainly in the certification: conjoint quantifies preference relative to other options, so it tells you what people prefer rather than how many units you’ll sell. It’s a preference read, and treating preference share as a sales projection is the most common way teams overreach with the method.
Where Maxdiff Fits, and Where Turf Fits
Conjoint is one of three trade-off tools, and matching the question to the right one makes every study count. The boundaries are clean once you see them.
MaxDiff prioritises a list of items. When you have 7 to 200 ideas, claims, benefits or product concepts and you need a defensible rank order with relative importance attached, MaxDiff shows small sets and asks for the best and worst in each, forcing the same kind of trade-offs conjoint does. It ranks interchangeable items on a single dimension. The design rule from aytm’s MaxDiff guide draws the line: when you’re testing one facet (which messages resonate, which flavours to prioritise), use MaxDiff; when multiple independent attributes interact, like feature and price and configuration, use choice-based conjoint. MaxDiff tells you which things matter most. Conjoint tells you how the things combine and what someone will pay.
TURF answers a third question: portfolio reach. Once you know what people prefer, TURF (Total Unduplicated Reach and Frequency) identifies the combination of items that appeals to the widest audience rather than the deepest. Its insight is that the most individually popular items often appeal to the same people, so a more niche addition can expand your base further than another crowd-pleaser. TURF is how you decide which three flavours to stock so the line reaches the most consumers, separate from which single flavour scores highest. It runs on ratings data or, notably, on MaxDiff output, which is why the three methods chain together in practice: MaxDiff to prioritise, conjoint to understand trade-offs and price and TURF to optimise the portfolio for reach.
Designing It Well
The method repays the teams that set it up honestly, and four practices from aytm’s certification do most of the work. Include a “None” option so respondents can walk away, because real purchase decisions include the choice to buy nothing. Anchor choices with a price, even when pricing isn’t the focus, because features sound free until they cost something, and a price reference grounds the responses. Prime respondents with an instruction screen up front, since the task is repetitive and clear expectations reduce fatigue. And let variables flow freely, so keep the combinations open instead of pre-restricting the ones you assume you’d never sell. Test everything, then use the market simulator to narrow to realistic scenarios afterwards, restricting only the genuinely impossible, like an iPhone running Android.
The Decision This Comes Down To
A product team deciding what to build and what to charge for it gets to choose how it decides. Run a conjoint, force the same trade-offs consumers make at the shelf and you get a quantified, simulatable read on which features drive choice and which attributes carry a price premium. That’s the answer to the question the team actually has: a clear, defensible picture of what to build, what to charge and why.
Explore the Choice-Based Conjoint Advanced Certification and the full Innovation Intelligence curriculum at the aytm Lighthouse Academy.
