Highlight Blog

Conjoint Analysis: Best Practices & How-To Guide

Written by Robin Kallsen | 7/1/25 2:56 PM

If you’d like to make your product more or less appealing to people in your target customer base, you’ve got plenty of options. You can lower the price, change the color, add a new feature … and so on.

But which changes will make the most difference in purchasing behavior? With such a wide range of options, testing them all seems a bit Herculean. Furthermore, everybody’s different; some people might always go for the cheaper item because they’re on a budget, for instance. And you need a way to understand purchasing behavior in the aggregate.

Thanks to the current heyday of market research innovation, you’re in luck. If someone was making one of those “The perfect market research method doesn’t exis—” TikToks, then conjoint analysis could very well be what pops up on the screen.

Why is that? Well, first off, it’s a method that’s specifically tailored to product testing research. Secondly, it’s designed to emulate how people make decisions out in the wild (a.k.a., the grocery store or the Amazon search results page)—times when they’re confronted with a variety of options and they have to make trade-offs among what’s available.

So, let’s dig into this research method and see what it can do for a product development team.

What is conjoint analysis?

Conjoint analysis is a market research method used to determine people’s purchasing behavior when presented with a number of different product variations.

Rather than asking people which specific features or attributes they want in, say, a pair of running shoes, a conjoint analysis survey will show them a few shoe profiles that differ from one another in specific ways. The survey-takers will need to say which shoe they’d be most likely to buy or rank the different shoe options by preference.   

The assumption here being that people don’t simply evaluate products on an attribute-by-attribute basis—it’s the combination that counts. This is more realistic than pretending that people are likely to evaluate each attribute separately when there’s multiple points of variation between products.

Even for the most data-driven of buyers—let’s imagine someone shopping on Amazon for essential oils who carefully scrutinizes the price per ounce—there’s really no way to completely isolate one variable. The essential oil decision will still depend on the brand, the look of the bottle, and whether or not there’s a dropper included.

How conjoint analysis works (without all the math)

Yes, there’s some fancy math involved in conjoint analysis market research (this is what software is for). No, you don’t need to understand multinomial logistic regression or Hierarchical Bayesian modeling to get a high-level feel for how to do conjoint analysis!

The building blocks of a conjoint analysis test are attributes (sometimes called factors) and levels. Attributes are things like color, price, flavor, and the like, whereas levels refer to the different options for each attribute (red or green, $30 or $50, vanilla bean or birthday cake).

For continuous variables like price, you’ll need to select either specific ranges or specific levels. Conjoint analysis for pricing works best when you first use the Van Westendorp Pricing Model to determine what price levels or ranges would be acceptable to most customers—particularly when a product is new and there’s not much pricing data to go on.

Once you have a list of your product attributes and the levels you want to offer for each, you put these together into multiple product profiles. A profile is a complete picture of a product in which every attribute has a designated level. Finally, you’ve got sets, which are combinations of profiles shown together.

By showing these sets to your testing audience and seeing which profiles “sell” best, you’ll learn two main things:

  1. the relative impact each attribute choice has on purchasing behavior (level part-worth)
  2. how much people care about each attribute in general (attribute part-worth)

If the above is confusing, imagine you’re at a bagel joint. The level part-worth for bagel type would be how much more you prefer sesame seed to jalapeno cheese, raisin, or pumpernickel. But what if you really don’t give a hoot about the bagels themselves, and you’re really just there for the lox and capers? Then you’d have a high attribute part-worth for toppings in general, a low attribute part-worth for bagel type, and very high part-worths for the topping levels corresponding to lox and capers.

The various flavors of conjoint analysis

Conjoint analysis research isn’t a one-size-fits-all solution. There are a few different methods depending on the nature of your product and what, exactly, you’re trying to find out.

  • Traditional Conjoint. In this type of survey conjoint analysis that dates back to the 70s and 80s, respondents would either rank or rate each product option presented to them. This can unfortunately lead to fatigue, particularly if you’re supposed to rank things that you feel very similarly about.
  • Choice-Based Conjoint (CBC). CBC is based on the idea that it makes more sense for respondents to select which option they’re most likely to purchase rather than try to rank all options. This is the type that most marketers mean when they say “conjoint analysis.”
  • Adaptive Choice-Based Conjoint (ACBC). Say you’re selling a Solve All My Problems device that comes in 3 different colors, 4 different patterns, 2 different sizes, 5 different scents, and 4 different textures. You can choose from 5 different built-in playlists, 3 different amounts of cloud storage, 4 different TV shows, and 11 different types of stardust that will shoot out on your birthday. Each device is optimized to solve one of 5 different illnesses, 6 different financial problems, 3 different mental health issues, 7 different career dilemmas … and that’s just the beginning. When the attribute list gets insane, ACA can help you simplify your analysis by zeroing in on just the attributes that matter most to each survey-taker and deprioritizing the rest. Kind of like social media algorithms, the (computerized) survey adapts to the respondent in real time.
  • Menu-Based Conjoint (MBC). Not too far off from CBC, MBC lets respondents select more than one profile option, like they were picking from a menu.

What about MaxDiff conjoint analysis? MaxDiff (or Best-Worst Scaling) is more of an alternative to conjoint analysis than a subcategory of it. If you’re just interested in comparing feature by feature, MaxDiff is simpler and probably more appropriate.

How to design a conjoint analysis study that’s truly meaningful

The first step in getting started with conjoint analysis is to understand your possible product variations and design the survey accordingly. Think about how the attributes and the levels within each attribute relate to each other.

For instance, you don’t want to treat a bunch of choices as mutually exclusive when they aren’t. If you’re selling burgers and there are four topping choices, it makes a big difference whether people can only choose one topping or if they can have as many as they like. In the former scenario, the toppings can simply be listed as levels under the “toppings” attribute. But the latter scenario is more complicated, and leads to two possible strategies: 1) create burger profiles with every possible combo showing up as levels under the toppings attribute, or 2) spread the topping choices out into a bunch of different “yes/no” attributes, such as “has bacon/doesn’t have bacon” and “has avocado/doesn’t have avocado.”

To avoid respondent fatigue, you’ll want to only show 3-4 profiles per set, and keep your survey conjoint analysis to a manageable length. Fortunately, today’s advanced software solutions are able to spread product variations evenly to all respondents, so that nobody has to consider everything (and you still get meaningful data).

Pitfalls and limitations of conjoint analysis

It’s important to know when to use conjoint analysis and when you might need something else to get the whole picture. As a very data-driven approach, it’s got a tendency to flatten out some of the more messy, irrational aspects of consumer behavior.

If your product is likely to be an impulse purchase, a conjoint study might not be the best choice since one major assumption is that shoppers are actively and rationally weighing trade-offs. Planning to advertise your product at a huge discount is an indicator that you’re appealing to impulsive behavior.

Conjoint analysis also assumes that there’s just one decision-maker for a purchase. In reality, a lot of products (like household items) involve multiple decision-makers, whose preferences might be in competition.

Finally, conjoint analysis doesn’t say anything about customer needs, or what’s specifically prompting someone to buy your product. For this, you’ll need other research methods like in-depth interviews and ethnographic studies.

How to strengthen conjoint analysis (Highlight can help here)

Having a large audience of committed survey-takers is a sine qua non. The more product attributes you’re testing, the more survey respondents are needed to evenly test all attributes without prompting survey fatigue.

Highlight can help you build testing audiences well into the hundreds, and we’ll make sure you get complete, authentic data with >90% survey completion rates. We can also help you segment your audiences to ultra-niche groups depending on your product testing needs.

If you’d like to delve into other types of market research, including ethnography, Highlight can help with this too.

A data-driven look at your product options

It’s time to drop the guesswork when you’re getting ready to settle on a few variations of your product. Conjoint analysis can give you clear indications of what people are really looking for in a product like yours.

And if you’re in need of a community of testers who will carefully consider each product profile set and respond authentically, you know where to find them. Right here at Highlight!