Predicting Revenue from Conjoint Results

Conjoint analysis is a powerful tool for predicting market reaction to changes in existing products or services or completely new products. But how well do conjoint model results translate into real world results? And can you predict revenue from conjoint results?

The input to a competitive conjoint model is the specifications for the current products/services in the market. The output is “shares of preference.” Shares of preference represent the predicted shares for the products in the model given equal awareness and equal distribution. As awareness and distribution are not equal in the real world, and as other effects, such as inertia, may exist, conjoint results do not reflect actual market shares.

From that starting point in the model, you run scenarios, which may change an existing product’s features or pricing or even test market preference for new products. Let’s say a product’s base share of preference is 10%, and when the price is dropped 25%, the share of preference increases to 15%. Can you translate that directly into a volume projection (50% increase in volume) and revenue projection (based on projected volume times the new price)? In a word, no. We like to think of the new share of preference as potential share. Actual share realized will differ for any number of reasons.

To get to a better volume and revenue estimate, a series of assumptions will need to be made. These might be different for every client and every industry. There’s not one right set of assumptions; there is subjectivity involved in translating conjoint shares of preference into predicted volume and revenue. There are a number of approaches we have used or have seen our clients use. We describe several of these below.

• Estimate actual awareness/distribution to be achieved. Since the model assumes perfect awareness and distribution, adjustments can be made to predicted shares of preference based on known or expected awareness and distribution for each product. This is often applied at the brand level. This approach may be especially appropriate for consumer packaged goods, where awareness and distribution are common measures and are typically known for each brand in the category.
• Track results over time and compare predicted to actual. Let’s say the conjoint model predicts that a new product will have 20% share of preference. After the product is introduced and some time has passed (perhaps one year), compare the 20% model prediction to the actual share achieved. This will result in an approximate “discount factor” to apply to future model runs.
• Model a historical action and compare to the model results. Let’s say your product was priced 10% higher one year ago, and everything else in the market was the same. The conjoint model can be run as if it were one year ago (with the higher price). Then it can be run at the current, lower price. The resulting predicted change in share of preference can then be compared to actual change in market share. As in the previous example, this will result in an approximate “discount factor” to apply to additional model runs.
• Cut any increase in share predicted by the model in half. We have several clients who, for lack of better information, simply cut any predicted increase in share in half. They consider this a conservative approach and have been pleased with the product and pricing decisions made as a result.

As you can see, there’s not just one approach to getting to predicted revenue, and the approach taken will depend on the industry and the amount of information available regarding competitive distribution, awareness and market share. While these methods may result in more accurate predictions, as with any market research, ultimately management judgment must be applied to a given business decision.

This content was provided by Sawtooth Technologies. Visit their website at www.sawtooth.com.