Every product, service or bundle accumulates features. Some were added for a big customer, some for a competitor comparison, some because an engineer thought they were elegant. Each one costs money to build, maintain, support and explain. Few are ever removed.
A Customer Value Audit answers a simple question: which of these do customers actually value, and what are they willing to pay for? Three methods do most of the work. This post explains what each one is for, where it fails, and how to combine them.
MaxDiff: what matters most and least
Maximum difference scaling shows respondents small sets of four or five items and asks which is most and least important. Across many sets, it produces a ratio-scaled ranking of every item on the list.
Use MaxDiff when:
- You have a long list, typically 15 to 40 features, benefits or messages, and need to know their relative importance.
- You want a ranking that is robust to the usual survey problem of everything being rated "important".
- You need to see how priorities differ between customer segments.
MaxDiff is fast, cheap and easy to explain to a management team. Its limit is that it ranks items in isolation. It tells you that feature A matters more than feature B. It does not tell you whether customers would pay for A, or whether they would accept losing B if the price fell.
Conjoint analysis: trade-offs and willingness to pay
Conjoint shows respondents complete product profiles, each a combination of attribute levels including price, and asks them to choose. From the choices, a model estimates the value of each level and the trade-offs people make between them.
Use conjoint when:
- You need willingness-to-pay estimates for specific features or service levels.
- You want to find the strongest combinations, not just the strongest single items.
- You need a simulator that lets the product or pricing team test configurations and see predicted share and revenue.
- You are deciding between bundles, tiers or price points.
Conjoint is more demanding. It needs careful design of attributes and levels, a realistic choice task, and a sample of real buyers. Six to eight attributes is the practical ceiling for one exercise. Push beyond that and respondents simplify, and the data with them.
The output that pays for itself is the simulator. A management team can sit with it and ask: if we remove these two features and drop the price by eight per cent, what happens to preference in the mid-market segment? The answer is a model estimate, not a fact, but it is a far better basis for a bet than opinion.
A/B testing: does it change real behaviour?
MaxDiff and conjoint measure stated preference under controlled conditions. A/B testing measures what people actually do when shown different configurations, messages or offers in the real product or sales process.
Use A/B tests when:
- You have narrowed the options to a few concrete alternatives.
- You have enough volume to reach statistical significance in a reasonable time.
- The change can be implemented cheaply in a test form, such as a pricing page, an onboarding flow, or a proposal template.
A/B tests are the only method that proves behaviour. They are also the only method that cannot tell you why, and they can only compare things you have already built. Running A/B tests without prior research usually means testing the wrong things, one pair at a time, for months.
How the three fit together
The sequence that works in practice:
- MaxDiff to sort a long feature list into what matters, what does not, and how that varies by segment. This removes the bottom of the list from further consideration.
- Conjoint on the remaining attributes to measure trade-offs, willingness to pay and the strongest combinations, and to build a simulator.
- A/B tests on the two or three configurations the simulator says are strongest, to confirm the effect on real behaviour before a full rollout.
Not every audit needs all three. A service business with a short feature list may go straight to conjoint. A digital product with high traffic may rely more on testing. But the logic holds: reduce the list, measure the trade-offs, validate in market.
What comes out of a Customer Value Audit
The deliverables are practical rather than academic:
- Which attributes customers value most and least, by segment.
- Willingness-to-pay estimates for the features that carry value.
- Which combinations create the strongest proposition.
- Which elements can be simplified or removed without damaging demand.
- Recommended product, service or bundle configurations.
- A simulator or decision tool where the trade-offs are complex enough to warrant one.
- Clear recommendations on what to strengthen, simplify, test or remove.
A data-led menu optimisation built on exactly this logic increased annual revenue by SEK 34 million within one year. The gain came as much from removing low-value items as from adding anything.
Common mistakes
- Asking about features in the abstract instead of in realistic combinations with a price attached.
- Using a general consumer panel for a B2B decision.
- Treating the average as the answer when segments disagree.
- Skipping the simulator and delivering a report nobody can interrogate.
- Removing a feature the research says is low value without checking whether a small, profitable segment depends on it.