Learn more about Conjoint Research Tests in the Lighthouse Academy!

In This Article

• Building a Conjoint Test
• Selecting a Methodology
• How Many Completes Do I Need?
  • Conjoint Segmentation (recommended)
  • Conjoint Express
• Best Practices in Conjoint Testing

Building a Conjoint Test

1. Click the Conjoint icon in the left sidebar, drag it into place, or select Conjoint from the + Add a Question menu at the bottom of the Survey Editor.
2. Type or paste question text into the Question field.
3. Click the dropdown to the right of the Question field to select Express or Segmentation mode.
4. Type or paste up to 7 items into the Attributes/Features fields.
5. Type or paste up to 25 items into the relative Attribute option fields.
6. Check the box next to N/A option to allow respondents to select none of the packages on screen for each task.
7. Click the packages per screen dropdown to determine how many packages each respondent will see per screen.

Selecting a Methodology

Learn more about Conjoint Express methodology and Conjoint Segmentation methodology.

How Many Completes Do I Need?

400 completes is a hard minimum for any Conjoint test—you can't launch a survey below it, and the Results page flags your data as directional until you reach it. This floor applies to both methodologies, and new tests default to exactly 400 respondents.

Above 400, the right sample size depends on which methodology you chose in Selecting a Methodology. Express and Segmentation analyze data differently, which means their sample size requirements can vary based on your research objectives and the level of detail you're looking to achieve.

Conjoint Segmentation (recommended)

Segmentation is the right choice for most studies. It estimates utilities at the individual respondent level (using Hierarchical Bayesian modeling) and auto-discovers personas from the data. That individual-level detail is what lets you break results out by subgroup and uncover segments—and it's the reason sample size matters more here.

Your sample should scale with the size and complexity of your design. The more options an attribute has—and the more attributes you test—the more choices the model needs before it can estimate each option's value with confidence. A widely used rule of thumb for main-effects Hierarchical Bayesian modeling analysis gives you a rough statistical floor:

completes ≥ 500 × c ÷ (t × a)

• c = most levels/options on any one attribute
• t = screens each respondent sees—shown under the test as "We'll show N screens to each respondent"
• a = packages per screen (2 to 5, default 3), not counting the N/A option

Use the table below as a practical starting point for your design:

⚠️ Note: Segmentation shows each respondent about twice as many screens as Express, so plan to start around 700 completes and go up from there.

💡 Tip: The unique packages and possible packages counts shown under the test tell you how big your design has grown. As you add attributes and options, the number of possible packages climbs fast, and the editor shows a live "Combinations left" warning as you approach the limit. Very large designs are capped by processing feasibility rather than a single fixed number, so watch that counter instead of aiming for a specific total.

Comparing subgroups

Because Segmentation gives every respondent their own utilities, you can slice results by any subgroup included in your survey or any aytm-included traits (age, region, quota groups, etc.). When you filter Results to a subgroup, aytm doesn't re-run the analysis—it reports preference shares from the individual-level utilities already estimated. However, aytm won't assess model-fit quality below 100 respondents per group, so thin subgroups simply may not have enough respondent-level data to estimate stable utilities in the first place.

Keep these habits in mind when you have subgroups of interest:

• Size from the smallest group up, never from the total down. Whatever your smallest reportable group needs sets the size of the whole order.
• Account for incidence. If a group is only part of your population, back into the total you need:
  total = target per subgroup ÷ incidence of smallest group.

💡 Tip: Four evenly sized groups at 300 each means about 1,200 completes. But if your smallest group is only 15% of the population and you want 300 there, order 300 ÷ 0.15 = 2,000 to hit that target naturally—or use quotas to guarantee the count.

Since aytm's Conjoint Segmentation automatically discovers personas, there is no need to size those before fielding. However, more total sample yields cleaner, better-separated segments.

💡 Tip: We understand that you may need to balance sample costs with these recommendations. If the guidelines shown here do not align with your research budget, please reach out to your Account Manager for a consultation with our Research Team. We can help guide you to the optimal sample size for your specific conjoint design and study objectives.

Conjoint Express (lighter, total-level only)

Express is a faster, lighter option. It fits a single aggregate model across your entire sample—one set of utilities for everyone—rather than estimating each respondent individually. That makes it quick to run and more forgiving of larger designs, but it comes with real trade-offs.

When you filter Results to a subgroup in Express, aytm re-runs the analysis on just those respondents to generate a subgroup-specific model. This means a thin subgroup produces unreliable or directional results—not enough data to build a stable aggregate. Express also can't discover personas (no individual-level utilities to cluster), so you're limited to comparing pre-defined groups only.

Because the analysis runs at the level you report on, the 400 minimum generally covers a straightforward Express read at total sample—the design-complexity scaling above is Segmentation-specific.

⚠️ Note: For subgroup comparisons, individual-level insight, or persona discovery, choose Conjoint Segmentation instead.

When to use Express: a quick, directional read on a simpler design where one total-sample result is all you need.

Best Practices in Conjoint Testing

• Not sure how many completes to order? See How Many Completes Do I Need? above—400 is the enforced minimum, but the right number depends on whether you're using Segmentation or Express and, for Segmentation, on your design and the groups you plan to compare.
• As you add combinations and change the number of packages per screen, keep an eye on the message at the bottom of the test that shows how many questions the test is using and how many packages each respondent will see.
• To help combat respondent fatigue, add some Instruction Text before the Conjoint test to give respondents context and let them know what's coming next (e.g., "You will repeat this task 8 times."). Adding more screens is one way to satisfy the sample-size math above, but screens hit a fatigue ceiling fast—past a point, ordering more completes is the safer lever.