How Does Conjoint Analysis Actually Work?
Conjoint Analysis is one of the most powerful research techniques when it comes to pricing analytics, particularly optimizing price points, but especially when a company needs to understand the ‘value’ customers attach to a product or product feature.
Conjoint analysis starts with an exercise, in which survey takers make selections among various products evaluating the features and benefits of various products or services presented to them.
Because of this, Conjoint Analysis is often referred to as ‘trade-off’ analysis.
What is Conjoint Analysis Used For?
Conjoint Analysis is also very effective when looking to understand a product’s price elasticity, in finding the most optimal price point, in understanding what customers are willing to pay for a new service or a new feature on a product, in estimating or forecasting a new product’s demand and in understanding how valuable a brand-name is.
So, let’s take a deeper dive on how conjoint analysis works.
How Does Conjoint Analysis Work?
Conjoint analysis is a survey-based research technique — that means you need to find a sample survey takers that represent your target population. There are large national sample providers, such as Dynata, Cint, and many others that can help with that.
The data, which is used for the conjoint analysis comes from an exercise done by survey takers called the choice task. These survey takers are part of a sample that represents potential buyers in the market.
This here below is a choice task example, in which survey takers are asked to compare three credit cards.
They’re given relevant information about the three credit cards: what brand issued the credit card, what type of card is it, whether there is an annual fee, if yes, how much, whether you have cash back, and many other relevant attributes.
The purpose of the choice task is to measure how customers tend to choose in a certain product category. What product attributes tend to be important when making choices?
When survey takers reviewed these, they select the one they’d prefer — and then, a set of new credit cards are shown — again with all the information.
And it goes on and on — each survey taker makes about 8 to 10 choices. The series of choice tasks is called ‘the choice exercise.’
So, how does the choice-task exercise allow us to estimate price elasticity, optimize product features, forecast demand and the many fantastic business answers we get — from Conjoint analysis?
How can we do all of that from simply having people pick from options?
The Mechanics of Conjoint Analysis
Here is how it works on a high level:
Let’s suppose the product we want to analyze — and the products in the category — have three main important attributes: it may be its brand, package size, and price.
For illustration, we’ll call the attributes: triangle, square and circle.
And each attribute can have different levels, just like brand can be different brands or size can be different sizes, let’s just say, our shapes can be different colors. These colors represent the different levels of the attributes.
In the choice task — a survey taker is shown three products size by size. While the survey taker sees them as three different products, for the purposes of the conjoint analysis it’s really just an almost random combinations of the three shapes’ colors — shown three times side by side. (see below)
Because Conjoint Analysis is a probabilistic methodology, after hundreds of people making altogether thousands of choices among the different combinations of the colors of the various shapes, the analysis will look for patterns in those thousands of choices.
It will look at patterns on which color in each shape makes it more or less likely for survey takers to select a product when it is shown. In fact, it will calculate a score for each shape’s color — indicating whether that color increases or decreases the probability that a product will be selected. This score is called the part-worth utility score.
In our example of shapes and colors (see image above) — among the shape of triangle, it looks like green was the most appealing, conjoint analysis calculated a utility score of 1.52, followed by yellow, then blue and then orange. That means, in the choice task, the products in which the triangle was green — it tended to be selected by survey takers more than when it was yellow, and quite a bit more than when it was blue or orange.
The Total Appeal of a Product is the Sum of its Parts
In conjoint analysis, we make the assumption that the total appeal of a product is the sum of the appeals of its parts.
If this assumption holds, that the part-worth utility scores for each attribute level, or in our case for each shape’s color will sum up to a total utility, then we can construct any product from these shapes and colors and we can sum up these part worth utilities to get the total utility of that product — think about it as its total appeal.
So, if we think about an actual product, say a credit card, the part-worth utility scores for its attributes (like brand, APR, introductory rate, rewards, etc…) will sum up to the total appeal, or total utility that product provides.
In fact, once we estimated the part worth utility scores, we can ‘construct a product’ from the many attributes and calculate its total utility. As a matter of fact, we can construct a competitive landscape of many products using the attributes and their levels.
Once we have the total utility score for the many products we ‘constructed,’ we can use a statistical formula and estimate what the probability is for any of these products to be selected in this set of products by customers. This is called the Preference Share, and by calculating preference share, we can estimate potential demand for a new product launch. (see formula below)
Once we have estimated the part worth utility scores for each product attribute level, and we can ‘construct products’ to estimate their preference shares, we can also create a simple preference share simulator. In this simulator, we can simply change for one of the products any of the attribute levels and see how preference share changes. What happens to the total appeal for a Wells Fargo credit card if it charges an annual fee vs no annual fee? What if they offer travel points? The simulator allows researchers to investigate what-if scenarios, whether for your product or for competitors.
With these ‘what-if scenarios’, we can also construct a ‘sensitivity curve.’ The curve describes how preference share for a product responds to changes in an attribute. This is most often used with pricing to see how sensitive a product’s appeal or demand is to changes in pricing. The price sensitivity curve is really considered the product’s price elasticity curve.
In the next blog post, I will be going much deeper into conjoint analysis. We’ll learn about the experimental design, which is the blueprint of the choice exercise, we’ll learn about the hierarchical Bayesian estimation technique and many more important aspects of the conjoint analysis. Please follow me here — and you’ll learn a lot more about this extremely useful market research technique.
