By Bryan Orme, CEO & President, Sawtooth Software, Inc.
First, What Is Conjoint Analysis?
Conjoint analysis is used in marketing research (and economics) surveys to gain deep insights into how people make choices when considering different products or services to buy.
Rather than directly ask respondents which features they prefer, what’s important to them, or how price-sensitive they are, conjoint analysis questions present realistic-looking market scenarios and simply ask people which product they would choose in each scenario.
We show respondents typically eight to twelve conjoint analysis questions (scenarios), where the product features systematically vary from question to question. On average it takes respondents 10 to 15 seconds per scenario to answer.
A respondent who is extremely price sensitive will usually pick the lower priced products across multiple scenarios. A respondent especially loyal to a particular brand will tend to pick that brand each time, unless perhaps it is shown at the highest price point or at the worst performance level. People have different feature preferences and price sensitivity, and conjoint analysis enables us to build a predictive model capturing a wide variety of consumer behaviors.
We combine the features (called attributes and levels) in a carefully designed format, such that the product concepts reflect a fair and balanced experiment. Each brand (or color or price) shows approximately an equal number of times, as well as approximately an equal number of times with every other feature in the study, allowing the researcher to tease out the independent effect of each feature on product choice using, typically, a multinomial logistic regression analysis (MNL).
Originally developed in the 1970s, conjoint analysis has gone on to become the gold standard approach for product feature and pricing optimization. The most common conjoint analysis approach is CBC (Choice-Based Conjoint) and Sawtooth Software is the world’s most respected conjoint analysis platform provider.
Question Layout Imitates How People Make Decisions in the Real World
Conjoint analysis works so well because the conjoint analysis questions tend to imitate the layout and information on brand, features, and price as presented in the real marketplace.
To the degree that the conjoint analysis questions reflect the available information and approximate layout of how products are seen and evaluated in the real world, the responses people provide are highly predictive of their actual purchase behavior.
Covers Much More Product Option/Price Space than Traditional Methods
A conjoint analysis survey with, say, 300 respondents can successfully predict preferences for thousands or millions of product variations (unique combinations of brand, color, performance, style, and price). Thus, conjoint analysis is much more powerful and flexible than traditional concept tests, A/B tests, Van Westendorp (Price Sensitivity Meter) tests, or Gabor-Granger pricing tests that can only cover one or a very few product variations.
Respondents Don’t Have to Evaluate All Product Options/Prices
Take, for example, a modestly sized conjoint analysis study with 6 attributes each with 4 levels per attribute. For example, there may be 4 brands, 4 colors, 4 prices, etc. across 6 total attributes. Mathematically, there are 4x4x4x4x4x4=4096 possible product combinations that can be constructed, for this example.
Conjoint analysis uses carefully designed experimental plans (sets of product concepts) that cover each attribute level multiple times. Luckily, the software handles the heavy lifting! In these balanced plans, each attribute level shows about an equal number of times with every other attribute level. Even though there may be 4096 total product concepts that can be built in our example above, each respondent may only need to evaluate 24 total product concepts (say, 8 choice tasks involving 3 products per choice task). Different respondents get different versions of the conjoint questions, such that across respondents a few hundred unique product concepts will have been evaluated.
Because conjoint analysis assumes an additive model, where the total value of a product is made up of the sum of the preferences (utilities) for its independent parts (its attribute levels), we don’t need to ask respondents to evaluate all possible combinations of the attributes and levels. An intuitive way to think about the scope of the problem we presented above is that, with 6 attributes with 4 levels each, we need to learn about 24 total preference values (utilities) rather than 4096. A subset of the total possible combinations therefore is sufficient to provide all the data we need to gain deep insights into preferences for these 24 levels and to make excellent predictions regarding which (among all 4096) products will win and lose in the marketplace. Additive combinations of these 24 levels allow us to rank-order the preferences for all 4096 possible products, for each respondent.
Needs-Based Segments Can Be Accurately Identified and Targeted
If you’ve interviewed enough respondents to contrast the unique preferences for groups of respondents (typically about n=200 per market segment or more), conjoint analysis is probably the best way to find and understand needs-based segments for products made up of multiple attributes. For example, one segment may be particularly interested in high-performance products at a relatively low price. Another segment may be interested in high-reputation brands with the most durability. You can then cross-tabulate segment membership with other descriptive variables you’ve collected for these respondents in your market, to figure out how to target them.
If you are creating multiple products to serve different segments of the market, conjoint analysis can help you optimize the products to grow your market share and increase your overall revenues and profitability.
The statistical approach used for finding these market segments in conjoint analysis data is typically Latent Class MNL.
The Market Simulator Is Flexible, Intuitive, and Powerful
The ability to deliver a what-if market simulator tool often in a spreadsheet file format (as shown below) is one of the reasons conjoint analysis has become so popular for marketing, strategy, and business decisions over the last few decades.
The market simulator lets the researcher or strategy manager specify and place multiple products in competition with one another, where the products are defined using one level from each of the attributes covered in the conjoint analysis study.
After the competing products have been specified in the market simulator, the shares of preference (similar to the concept of “market shares”) are rapidly calculated and shown for each product. The market simulator is like a voting machine, where each respondent in the data set “votes” on the product concept (out of the multiple products specified in the market scenario) they are most likely to choose. Fortunately, we don’t have to recontact the respondents again to ask them to vote: the preferences we extracted from their conjoint analysis questionnaire data are all we need to know to predict their preferences among thousands or millions of potential market scenarios we could specify in the market simulator.
The market simulator allows the researcher or strategy manager to test a variety of configurations and prices for their product (or product line). Search algorithms can be programmed into the market simulator, allowing the computer to do the heavy lifting to find products that optimize share, revenue, or profit for a single or multiple market segments.
Are There Drawbacks or Common Pitfalls?
Of course, conjoint analysis isn’t foolproof. Simpler problems can be done fairly easily, by skilled quantitative researchers using good software and with only a modest degree of training. Harder studies still can use the same good software, but typically need the oversight of an experienced conjoint analysis researcher.
Although it may sound counterintuitive, the hardest part of getting conjoint analysis right is developing the right list of attributes and levels. Programming the survey (using the point-and-click interface), designing the experiment, the utility estimation, and building the market simulator are the easier steps…for which good software takes care of most of the details and execution.
Interviewing the right people and incentivizing them to give realistic answers is paramount to success. It’s important to use multiple methods to identify bad respondents and clean (delete) them from the data set. Fortunately, conjoint analysis involves a fit statistic that reports how consistently each respondent has answered the conjoint questions. But cleaning the data involves more than just ensuring consistency in answering conjoint questions.
The market simulator predictions of “shares of preference” assume that all important attributes have been represented in the conjoint analysis study, equal information, full awareness of brands and products, and that each person is in the market to buy. The market simulator cannot automatically account for, say, the superiority of the promotional efforts or effectiveness of salespeople for one brand vs. another.
Where to Go to Learn More about Conjoint Analysis?
For decades, Sawtooth Software has been the recognized leader in the conjoint analysis platform space. Sawtooth Software’s Discover is an intuitive platform for composing, fielding, and analyzing conjoint analysis studies. Go to Discover.SawtoothSoftware.com to create a free account and start programming, fielding, and analyzing CBC-type conjoint analysis studies. The free version can be used for up to n=50 respondents per study.
ABOUT THE AUTHOR
Bryan Orme, CEO & President, Sawtooth Software, Inc.
Bryan is the recipient of the American Marketing Association’s 2017 Charles Coolidge Parlin Award, an honor reserved for those who “have demonstrated outstanding leadership and sustained impact on advancing the evolving profession of marketing research over an extended period of time.” He has published over eighty articles and white papers on conjoint analysis and received the David K. Hardin award for the best paper published in Marketing Research during 2004. He also authored the book Getting Started with Conjoint Analysis (now in its 4th edition) and co-authored the books Becoming an Expert in Conjoint Analysis and Applied MaxDiff. In his spare time, Bryan enjoys travel, hiking, and scuba diving.