By the Wavefield Research team · Published Oct 5, 2026
Conjoint analysis is a survey method that measures what people value by forcing trade-offs: respondents repeatedly choose between product options whose features and prices vary systematically, and the analysis converts those choices into a value score for every feature level. It is the heavyweight tool of product and pricing research — and the most over-prescribed.
Straight disclosure before anything else: Wavefield does not offer conjoint designs. This page explains how conjoint works, what running one properly involves, and the decision guide we genuinely use — because most teams who think they need a conjoint need something simpler, and some really do need the heavyweight tool.
You define a product as a set of attributes (price, content, screens, ads) and each attribute's levels ($7.99 / $9.99 / $12.99; with ads / without). The software generates product profiles that combine levels in a statistically designed pattern, and respondents face a dozen or so choice screens — a conjoint analysis example looks like the screen below, a streaming subscription with four attributes — each time picking the option they would actually buy, or none.
Option A
Price: $12.99 / month
Content: Full catalog
Screens: 1 screen
Ads: No ads
Option B
Price: $7.99 / month
Content: Full catalog
Screens: 4 screens
Ads: With ads
Option C
Price: $9.99 / month
Content: Limited catalog
Screens: 2 screens
Ads: No ads
No respondent ever says what matters to them — the choices reveal it. The analysis (hierarchical Bayes estimation, in practice) produces utilities: a numeric value for every level, per respondent. From utilities come the two outputs buyers care about: attribute importance scores (price drives 40% of the decision, ads 25%…) and share simulators — models that estimate how preference would split if you launched configuration X against competitors Y and Z. The method's canonical reference is Sawtooth Software's conjoint documentation, the industry-standard tooling for these designs.
Design expertise. Choosing attributes and levels is where conjoint studies are won or lost: miss the attribute customers actually trade on, or define levels nobody would recognize, and the elegant math optimizes a product in an imaginary market. Experienced conjoint designers earn their fees here, before a single respondent is sampled.
Sample and respondent burden. Plan on 300+ respondents per segment, each completing 10–15 choice screens — a cognitively demanding task where data quality discipline matters even more than in ordinary surveys. Specialized analysis follows: utilities, importance scores, and simulators come from dedicated conjoint analysis software, not a crosstab engine.
The design rules themselves are unforgiving. Six or seven attributes is the practical ceiling before choice screens overwhelm respondents, with two to five levels per attribute. Levels must be realistic (prices customers would actually see), balanced in range (a deliberately absurd level inflates its attribute's apparent importance), and mutually plausible in combination — the design needs prohibitions so respondents never evaluate a premium bundle at an economy price nobody would offer. Each of these choices quietly shapes the utilities that come out the other end, which is why the misreads section below matters as much as the mechanics.
Realistically, a properly run conjoint study is a five-figure project through an agency or a specialist in-house team. That price is justified when the decision it informs is a pricing architecture or a product line worth millions. It is badly spent when the real question was simpler — which is the next section.
Importance scores are artifacts of the ranges you tested. The most-quoted conjoint output — “price drives 40% of the decision” — depends entirely on how wide a price range the study included. Test $7.99 to $12.99 and price looks moderately important; test $7.99 to $29.99 and price dominates everything, same product, same customers. Importance scores compare the levels you chose to test, not the attributes in the abstract, so reading them across studies — or quoting them without the ranges attached — produces confident nonsense. Any report that leads with importance percentages and buries the level definitions deserves a second look.
Share simulators predict preference, not sales. The simulator assumes every respondent is in the market, knows all the options exist, and chooses purely on the attributes tested. Real markets have awareness gaps, distribution, habit, and brand effects the model never saw. Simulated share is a decision-support number for comparing configurations against each other — configuration A beats B by 12 points — not a forecast of quarter-one revenue. Teams that treat it as a forecast are the source of most conjoint-disappointment stories.
You need conjoint when the question is a trade-off with money attached. “Would customers pay $3 more for the tier with the extra feature?” “Which bundle maximizes revenue against these two competitors?” Quantified willingness-to-pay for specific configurations, and demand simulation across a product line — that is conjoint territory, and nothing simpler substitutes.
“Which features matter most?” is not a conjoint question. A ranking question handles five to seven items, and MaxDiff (best–worst scaling) handles twenty — both at a fraction of the cost, both producing a defensible priority order without price simulation you were never going to use.
“What price can we charge?” for a single offer is a Van Westendorp question. Four questions, any competent survey, a defensible acceptable-price range. It will not simulate feature trade-offs — but if you only have one configuration, there is nothing to trade.
“Which concept wins?” is a split-sample test. Randomized arms, one concept per arm, significance between arms — the design covered in concept testing, and the one Wavefield runs natively. Most “we should do a conjoint” conversations, pushed one question deeper, turn out to be one of these three simpler studies. The ones that don't — genuine configuration-and-price optimization — deserve the real thing, run on specialist tooling.
A survey method that figures out what people actually value by forcing trade-offs. Instead of asking 'how important is price?' (everyone says very), it shows respondents repeated choices between realistic product options whose features and prices vary systematically. From hundreds of those choices, the analysis computes how much each feature level drives preference — numbers you can use to simulate how a market would respond to a product you haven't launched yet.
The dominant form of conjoint, usually abbreviated CBC: respondents repeatedly pick one option from a set of full product profiles, mimicking a real shelf decision. Alternatives exist — adaptive conjoint adjusts questions per respondent, and menu-based conjoint handles configurable products — but when someone says conjoint today they almost always mean choice-based.
Plan on 300 or more per distinct segment you want to read, and more as attributes and levels grow. Conjoint's power comes from modeling, and the models need data: a thin sample produces confident-looking utilities with wide error bars. This is one of the reasons a proper conjoint study rarely comes in under five figures once design, sample, and analysis are counted.
Sawtooth Software is the industry standard for designing and analyzing conjoint studies; general survey platforms mostly do not support real conjoint designs, and Wavefield is honest about being one of them — we don't offer conjoint. If your question genuinely needs conjoint, use a specialist tool or a research agency that runs one. If your question is simpler than conjoint (most are), the methods on this page answer it for a fraction of the cost.
MaxDiff (best-worst scaling) ranks a list of items by repeatedly asking which is best and worst in small sets — it produces a priority order, nothing more. Conjoint evaluates whole product configurations and puts price in the trade-off, producing willingness-to-pay and share simulation. The practical rule: a list of features to prioritize is MaxDiff; features bundled with prices into products you might actually launch is conjoint. MaxDiff is dramatically cheaper and answers the question most teams actually have.
Run properly — specialist design, 300+ respondents per segment, licensed conjoint software, analyst time for estimation and simulators — a commercial conjoint study typically lands in the $15,000 to $50,000+ range through an agency, with complex multi-segment studies going higher. DIY tools can cut that, but the savings concentrate exactly where studies fail: attribute selection and level design. Budget pressure is itself a signal to check the decision guide above, because the simpler methods answer most questions for a few hundred dollars.
Van Westendorp measures price acceptability for a product as a whole: four questions produce a range of acceptable prices. Conjoint measures how price trades off against specific features and can simulate demand at different price points for different configurations. If you need a defensible price range for one offer, Van Westendorp is faster and far cheaper; if you need to know whether customers would pay $3 more for the premium tier with the extra feature, that trade-off question is conjoint territory.
Related: Van Westendorp pricing · concept testing · ranking survey questions · market research methods
Feature priorities, price ranges, concept tests with randomized arms — briefed, built, and fielded in days, with significance testing built in. From $99 per study.
Start a study