5 Conjoint Analysis Examples — And the One That Cost a Launch
Conjoint is the most powerful stated-preference method there is. These examples show how it works — and the moment it breaks, because nobody actually buys inside a survey grid.
You came here for examples, so let's start with a clean definition and then get straight to five of them. Conjoint analysis shows respondents a series of product profiles — each one a bundle of features plus a price — and asks them to choose between profiles across a handful of choice tasks. Because price is varied alongside everything else, the statistics can back out the part-worth utility of each attribute level: how much each feature, brand, or price point adds to (or subtracts from) preference. Convert those utilities into money and you get implied willingness-to-pay. That's the whole engine. Now the examples.
Five worked conjoint analysis examples
Example 1 — SaaS pricing tiers
A B2B software team wants to know which features justify a higher plan. Attributes and levels:
- Seats: 5 / 20 / unlimited
- SSO & admin controls: no / yes
- API access: no / yes
- Support: email / priority / dedicated CSM
- Price/month: $49 / $99 / $199 / $399
Respondents pick between plan profiles. The output is a part-worth for every level — and dividing each feature's utility by the price-utility slope gives a dollar value. Say SSO is worth ~$60/month of utility and dedicated CSM ~$110. The decision: bundle SSO into the mid-tier (cheap to grant, high pull), and reserve the CSM for the top tier where buyers will actually pay for it.
Example 2 — Consumer electronics launch
A hardware brand is finalizing a phone and must trade screen against battery against price. Attributes:
- Screen: 6.1" / 6.7"
- Battery: 3,500 / 4,500 / 5,500 mAh
- Camera: dual / triple
- Price: $599 / $699 / $799 / $899
The output is a willingness-to-pay map: an extra hour of battery might be worth ~$45, while the larger screen is worth only ~$20. The decision: spend the bill-of-materials budget on the bigger battery, not the bigger panel — and price the flagship at $799 rather than $899, where the part-worth for price falls off a cliff.
Example 3 — FMCG pack and price
A packaged-goods team is choosing a pack size, an on-pack claim, and a shelf price. Attributes:
- Pack size: 250g / 500g / 1kg
- Claim: none / "high protein" / "no added sugar"
- Price: $3.49 / $3.99 / $4.49 / $4.99
The output is a simple share-of-preference simulation: define your candidate product and two competitor profiles, and the model estimates the percentage of choices each captures. The "no added sugar" claim might lift share from 31% to 38% at the same price. The decision: lead with that claim and hold the $3.99 line, because the simulation shows you lose more share at $4.49 than the extra margin recovers.
Example 4 — Brand vs. price
How much is brand equity actually worth in dollars? Put brand in the design as an attribute:
- Brand: Your brand / Premium rival / Private label
- Feature set: standard / premium
- Price: $20 / $30 / $40 / $50
The output: the brand attribute carries its own part-worth. If the gap between your brand and the private label equals the part-worth of a $12 price difference, your brand is "worth" a $12 premium per unit. The decision: you now have a defensible price premium to anchor on — and a number to defend in the room when finance asks why you're not matching private-label pricing.
Example 5 — Roadmap feature prioritization
Engineering can build three of eight candidate features this year. Run a conjoint with the strongest candidates as attributes (each present/absent), plus price as the anchor. The output is a ranked list of part-worths — and, crucially, a share-of-preference simulator. You can ask: "If we ship feature A and C, what's our preference share against today's product?" versus "A and D?" The decision: sequence the roadmap by the combination that moves share most per engineering-week, not by whoever lobbied loudest in planning.
The example where it failed
Here's the one that should make you cautious. (Illustrative — not a named client.) A premium kitchen-appliance maker ran a textbook conjoint before launching a flagship product. The study was clean: good design, large sample, tight confidence intervals. It said the target customer would happily pay $1,200.
The team got nervous and launched at $899 — well under what conjoint promised. Demand was still weak. Returns were high. The forecast and the market told completely different stories.
Why? Because the survey grid removed the three things that govern a real purchase. It removed real budget — in the survey, choosing the $1,200 option costs nothing. It removed real alternatives — the full, messy choice set of everything else a household could spend $1,200 on. And it removed real consequence — there's no buyer's remorse for a click. What conjoint measured was a considered, articulate, slightly aspirational version of the customer. What showed up at checkout was the actual one. This is the say-do gap in its most expensive form: stated preference and revealed preference are not the same currency, and in premium categories the gap runs in your face, not your favor.
"Conjoint doesn't lie. It answers exactly the question you asked — which is 'what would you choose if nothing were at stake?' That's just not the question your P&L is asking."
When conjoint is trustworthy — and when it isn't
Conjoint isn't broken. It's misapplied. Here's the line.
| Your Situation | Conjoint Verdict |
|---|---|
| Low-involvement category, everyday purchase | Trustworthy — buyers decide fast in real life too |
| Many features, you need a clear hierarchy | Trustworthy — this is conjoint's home turf |
| Directional read on what to build first | Trustworthy — relative ranking is robust |
| Premium / luxury positioning | Validate — stated WTP runs optimistic |
| Strategic price increase or decrease | Validate — needs a real demand-response model |
| Emotional or identity-driven category | Validate — articulated reasons ≠ real drivers |
How to fix it
The fix isn't to throw conjoint away. It's to stop treating it as the final answer for high-stakes decisions and start treating it as one input. Two moves:
First, validate against behavioral and implicit data. Stated trade-offs tell you what people will articulate; implicit measurement captures the gut-level acceptance that actually predicts checkout behavior, especially in premium and emotional categories where the say-do gap is widest.
Second, model the real demand response with Causal AI. Conjoint gives you part-worths under survey conditions. Causal AI works from real market and behavioral data to estimate what would actually happen to demand if you moved the price — the causal question, not the correlational one. At SUPRA this is the standard pattern: use conjoint to map the option space cheaply, then validate the load-bearing numbers with behavioral data and a causal demand model before anyone commits a launch price. Cheap exploration, validated commitment.
Frequently asked questions
What is a simple conjoint analysis example?
Ask people to choose between phones described by three attributes — screen size, battery life, and price. Each respondent picks from several profiles across a handful of choice tasks. The statistics then reveal how much each attribute level adds to preference (its part-worth utility) and the implied dollar value of, say, an extra hour of battery.
What is conjoint analysis used for?
To quantify how much customers value individual product attributes and to estimate willingness-to-pay for features. Common applications: pricing tiers for SaaS, feature trade-offs in product launches, pack-and-price decisions in FMCG, valuing brand equity in dollars, and prioritizing a roadmap by share-of-preference simulation.
How many attributes should a conjoint study have?
Usually four to six attributes with two to five levels each. Fewer than four and the model is too coarse; more than six and respondents fatigue, making trade-offs unreliable. If you have many features to rank, MaxDiff is a better fit than overloading a conjoint design.
Does conjoint analysis predict real purchases?
Sometimes. It predicts well for low-involvement, multi-feature categories where decisions are directional. It predicts poorly for premium positioning, strategic price moves, and emotional categories, because the survey grid removes real budget, real alternatives, and real consequence. Validate high-stakes conjoint output against behavioral data and Causal AI before betting a launch on it.
Is your conjoint telling you the truth?
Most clients come to us because a clean conjoint study said one price and the market did another. We pressure-test the numbers that carry your launch — before you commit to them.
Get my AI Diagnostic →