A worked example in two waves: Anchored MaxDiff finds the features that are worth money, CBC puts a price on them, and Van Westendorp cross-checks the price range.
New SaaS products appear every week, and most of them face the same questions before launch: what goes into which plan, what to charge, and how to bill. This guide walks through one such product from the first feature list to a priced plan.
An AI agent that creates software tutorials. You enter the address of your web app and describe a workflow, for example “show how to create a project”. The agent clicks through your app in a real browser and delivers a finished tutorial video with voiceover.
The founders want to know:
Fourteen candidate features are too many for a conjoint study: respondents cannot weigh fourteen attributes at once, and the answers get noisy. Asking for the price of each feature directly does not work either, because stated prices for single features say little about what people choose. The study therefore runs in two waves.
Wave 1 screens the features. An Anchored MaxDiff ranks all 14 features and shows which ones are worth paying for. A Van Westendorp block adds a first acceptable price range.
Wave 2 prices the features that matter. A CBC combines price, contract term and video minutes with the four strongest features from wave 1. The market simulator turns the choices into willingness to pay, a demand curve and the right annual discount.
Each wave is its own survey. Wave 2 is built after wave 1 has been analysed, because its attributes depend on the wave 1 results. Choicery does not pick attributes per respondent within a single survey, so the selection happens between the two waves and is the same for everyone in wave 2.
Two single-choice questions come first: company size (just me, 2 to 10, 11 to 50, 51 to 250, more than 250) and where the company is based (Germany, another EU country, outside the EU). They cost a few seconds and later split every result into segments. The region matters most for the language features: a sample of German companies will put German voiceover first almost by definition.
Each feature is one short statement. English voiceover is always included, and the instruction says so, so nobody spends a choice on it.
| # | Feature |
|---|---|
| 1 | Automatically re-records the video after every release of your software |
| 2 | Voiceover also in German |
| 3 | Voiceover also in French |
| 4 | Voiceover also in Spanish |
| 5 | Your own voice as the narrator |
| 6 | AI avatar as presenter |
| 7 | Step-by-step written guide with screenshots |
| 8 | Interactive click-through demo |
| 9 | Your own branding (logo, colours, intro and outro) |
| 10 | Sensitive data is blurred automatically |
| 11 | Embed the videos in your help center or docs |
| 12 | Viewer statistics: who watched and where they stopped |
| 13 | Edit the script before the video is created |
| 14 | Data is processed in the EU |
Each screen shows 4 features, and the respondent picks the most and the least important one. With 14 features and 4 per screen, 11 screens show every feature about three times (3 × 14 / 4 ≈ 11), the usual rule of thumb from the MaxDiff guide. The three language features are deliberately separate items: the ranking then shows which language to add first, instead of one vague “more languages”.
Price is not one of the features. “Low price” has no defined level, almost everyone would pick it as most important, and MaxDiff cannot say how much money a feature is worth. Price enters in wave 2, with concrete amounts.
A plain MaxDiff only ranks the features against each other. Some feature always comes out on top, even if none of them would change a purchase. The anchor fixes that. After the last screen, respondents answer one yes-or-no question for every feature:
“Would you pay more for this feature?” Yes, I would pay more / No.
This is the direct binary anchor (Lattery, 2010). It places a threshold on the same scale as the MaxDiff scores: features above it are worth money to the respondent, features below it are not. The wording ties every feature to willingness to pay without naming an amount. The amount comes from the CBC in wave 2. How Choicery estimates the anchored scores is described in the Anchored MaxDiff guide.
The last block asks the four Van Westendorp questions (Van Westendorp, 1976) about one fixed plan:
The Pro plan: 60 video minutes per month, automatic re-recording after every release, voiceover in English and German, and a step-by-step guide with every video.
The result is an acceptable price range from PMC to PME. Its main job in this study is practical: it tells you where to put the price levels of the CBC in wave 2, so that the levels cover what people find acceptable instead of a range picked from competitors’ price lists.
The CBC in wave 2 has seven attributes. Three are fixed in advance, because they are the pricing questions themselves:
The other four come from wave 1. Four rules guide the choice:
Above the anchor. Only features that a relevant share of respondents would pay more for are candidates. A feature below the anchor goes into the cheapest plan or is not built.
Reach, not just rank. A TURF analysis on the MaxDiff scores shows which combination of features appeals to the most respondents. Four features that all appeal to the same people add less than four that cover different groups.
Expected features go into every plan. A feature that nearly everyone wants and every competitor offers has nothing to trade off. It belongs in all plans, not in the CBC.
Related features become one attribute. The language items turn into one attribute with three levels, and the written guide and the interactive demo into one attribute for what the customer gets. This keeps the CBC at seven attributes.
Every task shows three plans and the option “I would not subscribe to any of these”. Respondents answer 10 tasks. The None option matters for pricing: without it, the study can only say which plan wins, not whether anyone would buy at that price (see the reservation price guide).
| Attribute | Levels |
|---|---|
| Price per month | €19 · €34 · €49 · €64 · €79 |
| Contract | Monthly, cancel anytime · 12 months, billed monthly · 12 months, paid upfront |
| Video minutes per month | 15 · 60 · Unlimited |
| When your software changes | You re-record manually · Re-recorded automatically after every release |
| Voiceover languages | English · English and German · English, German, French and Spanish |
| What you get | Video · Video and step-by-step guide · Video, guide and interactive demo |
| Data processing | USA · EU · Not specified |
The price levels are equally spaced, so price can be estimated as one linear effect. That is what the willingness-to-pay and reservation-price calculations need (see the willingness to pay guide).
Because price and contract term vary independently, the study measures what flexibility is worth on its own. The difference in willingness to pay between “monthly, cancel anytime” and “12 months, paid upfront” is the discount an annual contract has to offer before customers prefer it. A “two months free” offer can then be checked against the data instead of copied from other price lists.
Many providers say nothing about where data is processed. The third level measures what that silence costs compared with naming the EU or the USA.
Language features rank by where respondents work. Recruit the markets you plan to sell in, and read the language results per region.
Respondents do not pay for the plan they pick in a survey. Treat willingness to pay as a reference point for the price decision, and check it against real sign-ups after launch.
Wave 2 prices the features that wave 1 found relevant. If the two waves reach different people, the selection may not fit the second sample. Recruit both waves from the same population.
The acceptable range describes price perception for one fixed plan. It is not a revenue or profit optimum, and it does not replace the demand curve from wave 2.
Chrzan, K., & Orme, B. K. (2019). Applied MaxDiff: A practitioner's guide to best-worst scaling.
Lattery, K. (2010). Anchoring maximum difference scaling against a threshold: Dual response and direct binary responses. Conference paper (pp. 91-106).
Louviere, J. J., Flynn, T. N., & Marley, A. A. J. (2015). Best-worst scaling: Theory, methods and applications. Cambridge: Cambridge University Press.
Orme, B. K. (2020). Getting started with conjoint analysis: Strategies for product design and pricing research (4th ed.). Manhattan Beach, CA: Research Publishers LLC.
Van Westendorp, P. H. (1976). NSS Price Sensitivity Meter (PSM): A new approach to study consumer perception of prices. Proceedings of the 29th ESOMAR Congress, Venice.