Ayush Jain Team J

Conjoint analysis offers designers the opportunity to gather a considerable amount of valuable information and insight. The ability to determine what attributes and at what level customers prefer. Notice that I did not say which attributes they ‘prefer to buy’. With conjoint analysis we can partially bridge the gap between what people say they would do, and what they actually do. We are attempting to determine purchase preferences in a non-purchase situation. In doing so, we can obtain some very valuable information. We could take this a step further by monitoring multiple offers in the marketplace, in real time (such as Capitol One’s ability to mail out variation of a credit card offer and determine through completed applications who buys what) or by measuring tacit knowledge with predictive market technology. Neither of these are appropriate options for future offerings – and both can require significantly more resources than conjoint analysis.

The ability to determine which attributes, for which customers, at which levels allows us to design products that can be mass produce, but come a little closer to custom offerings. The benefit of designing products the match customer’s preferences should be obvious.

Conjoint can also tell us what level of utility an attribute has at a given level. Utility is a measure of worth relative to the other attributes. If we include price of the product as one of the variables in the study we can interpolate a dollar value for the attribute and in some cases determine price elasticity. Knowing the utility of a product’s features can help us to match the material and manufacturing costs to the price customers are willing to pay. 

The third benefit that can be derived from conjoint is the ability to determine market segmentation. Which customers will show maximum interest in the product at the given attributes levels? At the consumer level segmentation is the aggregation of customers based upon desired attributes of the product. And since conjoint analysis is all about determining attribute preferences it is a logical tool for this use. By tracking who completes the study we can do some basic calculations to determine subsets of the market and their specific preferences. If sampling is reliable, the results can be generalized for a larger population.

A conjoint study: What the respondent sees.

The people participating in your survey (the respondents) will see a selection of product offerings called choice attributes. The collection of attributes, each represented at a specific level is called a profile. The selection of attributes is extremely important and will greatly determine the type of analysis you end up with. Profiles can include the full range of critical attributes and is called full profile design. A subset of attributes, those that you are particularly interested in can be shown in a partial profile study. 

The more attributes you include, the more complex the survey, the more complex the calculations become, and the less reliable the analysis will be. That is not to say that you should not include every attribute that is important, but be aware of the cost. Conducting a second study in order to measure an omitted attribute would certainly be costly as well.

Typically the participant has a choice of two offerings with differing attribute levels. For instance we might ask them to first choose between these options:

Option one	Option two
Visa	Mastercard
$105 annual fee	No annual fee
11% interest rate	6% interest rate

The respondent picks the one that has the most appeal to them.

Profiles can be presented in pairs as shown above (Adaptive Conjoint Analysis or ADA) or in a series of options (Choice-based Conjoint or CBC). The choice can be as simple as choosing the preferred profile or can include a range of preferences such as a scale from 1 to 10. In either case an optional selection for ‘none’ can be included.

In a full profile conjoint study there are often too many options to test each and every one. In the survey above there are three attributes included in the profiles. If each of those attributes has three possible choices then there are 27 possible combinations. That is probably not too bad, but suppose that you are interested in the features of a complex consumer electronic device. It may have 50 or more attributes with nearly as many options for each. That would be a long a complex survey. As set of 12 attributes with 2 to 5 options each... say a total of 35 levels would reveal 186,624 possible combinations (Green, 1999). Few people have the time or patients to sit though that kind of study. Fortunately we can make use of orthogonal arrays (Addelman 1962) to reduce the number of profile we show the respondent. An array that includes only a fraction of the possible combinations can be used to accurately estimate utility for all attribute level main effects. This is called a partial factorial design conjoint analysis. Unfortunately if you are interested in interaction effects between the attribute levels a full factorial design will be necessary. 

Steps for designing a conjoint analysis study:

Step one: Determining the attributes

Each step of designing a study is critical, but none so important as determining the attributes to be measured. Include irrelevant attributes and the complexity of the study increases exponentially. Omit important attributes and a critical opportunity is missed. So how might we choose the attributes?

First, if there is existing data that measures customer preferences it should certainly be considered. Any attributes that the customer consistently shows indifference for would be likely candidates for omissions. Second, there is considerable tacit knowledge held by sales force, customer service, tech support and often executives within the company or distribution channel. This is certainly an area for concern as conjoint analysis is a technique best used for customer preferences. The preferences of management should probably not be part of the study. Lastly, new attributes that are under consideration are appropriate. Whether they offer a differentiating advantage, cost saving, or time savings, conjoint can be an excellent predictor of consumer preferences for future offerings.

Step two: At what levels should we measure these attributes.

Too many levels and the complexity grows – to narrow or ‘reasonable’ a selection and we may miss an opportunity.

Step three: Determining who will participate in the study

Obviously, in any study, the more participants you get involved, the more reliable the results. Depending upon your goals, you may want to work diligently to verify your study, validate you sampling and spread you research across a wide range of segments. These are critical in assessing the statistical significance of any quantitative research.

Step four: Determine the length of the study and how it will be administered

Knowledge of your target group will help to determine the value of their time and how much you should expect. Incentives can help, but the risk of an opt-out part way through the study increases with each additional question. Keep it to a realistic length, given your audience.

Though the study can be set up, calculated and distilled by hand, the use of an available software package may save considerable time. Computer administration has the added convenience of avoiding the entry of data by hand.

Step five: Analyze the data

Determine the value systems, utility and any other results that you set out to find. You may have collected data that can be mined beyond the goals and objective you set. Outcomes that result outside of the study’s objectives should be validated though additional testing.

Step six: Determine action items and take-aways

A formal debriefing session is always a smart step in the conclusion of a study. Determining what else might be of interest, what could have been done differently, how the outcomes will be realized and what the next steps should be are critical. Documentation of this process is extremely helpful when additional studies are undertaken.