Which Software Package Would Customers Choose?

The question we need to answer

Imagine a small software company preparing to launch a new product. The team must choose a price, a customer-support option, and a billing arrangement. People on the team have strong opinions, but opinions alone cannot show which combination customers would actually prefer.

A standard survey could ask customers whether price, support, or contract terms matter. Most people would probably say that all three matter. A conjoint survey asks a more useful question: when people must choose between realistic packages with different combinations of features, what tradeoffs do they make?

That is the value of this study. Instead of evaluating each feature in isolation, respondents make choices that resemble the decisions they would face in the real world.

1. Turn the decision into a simple design

We describe each software package using three attributes. An attribute is simply a feature that can vary. Each attribute has several levels, which are the specific options a respondent might see.

The survey randomly combines these levels into product profiles. On each of eight choice screens, a respondent sees two software packages and chooses the one they prefer. Random assignment lets us separate the effect of price from the effects of support and contract terms.

For example, a respondent might compare a $25 package with email support and annual billing against a $50 package with phone support and monthly billing. There is no obviously perfect answer. The choice reveals which tradeoffs feel worthwhile to that person.

2. Prepare the survey for real people

This example uses an illustrative sample of 300 respondents. Each person completes eight choice tasks, producing 2,400 total choices. A real study may need more or fewer respondents depending on the design, intended comparisons, available audience, and required precision.

Before sharing the survey, the researcher previews every screen on both a computer and a phone. They check that prices and features are easy to understand, that no impossible combinations appear, and that the choice task is not unnecessarily tiring. They also decide in advance which comparisons are most important and how incomplete or low-quality responses will be handled.

This preparation is not glamorous, but it is where a trustworthy study begins. Clear wording and a comfortable respondent experience usually matter more than adding extra complexity.

3. Read the example results

The table reports how each level changes the average probability that a package is chosen. Every estimate is compared with a reference level, which acts as the starting point for that attribute. Here, the references are $25, email support, and monthly billing.

A mean difference of -0.08 for the $50 price means that packages priced at $50 were about 8 percentage points less likely to be chosen than otherwise comparable packages priced at $25, on average. A value of +0.11 for phone support means that phone support increased choice probability by about 11 percentage points compared with email support.

The 95% interval shows the uncertainty around each estimate. Narrower intervals indicate greater precision. If an interval crosses zero, the data do not clearly distinguish an increase from a decrease at that confidence level.

4. Translate the numbers into a useful conclusion

The overall pattern is easy to explain. Higher prices make a package less attractive. Better access to support makes it more attractive. Annual billing creates a modest penalty compared with monthly billing.

The largest positive estimate belongs to phone support: an 11 percentage-point increase relative to email support. The largest negative estimate belongs to the $75 price: a 19 percentage-point decrease relative to $25. That suggests customers value stronger support, but the improvement may not fully offset a large price increase.

A practical recommendation might be: begin with a lower-priced monthly package, then test whether live chat provides enough added value without the operating cost of phone support. The best final decision would also consider revenue, support costs, market segments, and business strategy.

5. Know what the results do not prove

These figures are illustrative, and even real conjoint estimates require careful interpretation. An 11 percentage-point mean difference is not the same as an 11% increase in sales. It is also not a guaranteed market-share forecast. The result describes average choices within the study and the combinations respondents were shown.

Conjoint evidence becomes more useful when the sample represents the audience of interest, the profiles are realistic, and the analysis follows a plan established before looking at the results. Subgroup patterns can be explored, but small groups and many unplanned comparisons deserve extra caution.

6. Try the complete example yourself

You do not need to recreate this study from a blank page. Open the editable version to see the attributes, levels, choice tasks, and analysis workflow inside Conjoint Survey. Change the product, audience, wording, or features to match your own question.

The sample CSV lets you inspect the data structure before fielding a study. The codebook explains what each field means. Together, these files make it easier to understand the path from a respondent's choices to the final table.