Every product must have a value proposition, but not every customer values that proposition. It’s commonly useful to determine how customers value different components, features, or other attributes of a product. You can’t just directly ask “what would you be willing to pay?” – you have to derive it through analysis. As Jobs famously said “people don't know what they want until you show it to them”, and that’s what Conjoint analysis does. Choosing between alternatives uses less cognitive load, so we’re not as likely to abandon the survey or make something up.
Conjoint analysis works by breaking a product or service down into its component parts (referred to as attributes and levels), and then testing different combinations to identify what’s driving consumer preferences. Which phone do you prefer: $499 with a 6.5 inch screen, $299 with 19 hours battery life, or $399 but still has a headphone jack? Survey respondents choose from multiple combinations, then Multinomial Logistic Regression reveals the value of each item or ‘Part-Worth Utility’. Results are broken down by attribute and levels, so you can tell which screen size and battery life to maximize sales, and confidently drop the headphone jack like Apple did in 2017.
For Conjoint Analysis to work, the attributes you choose must be observable to the consumer. For example it makes no sense to ask consumers to choose between different quantities of ingredients in a perfume, as they often have no idea what goes into their favourite perfume. Instead you might ask them to choose between notes in the fragrance, like citrus or vanilla. The attributes must also be knowable to the majority of the people choosing. For example it makes no sense to ask someone choosing a meal at a restaurant whether they enjoy Tarragon or Parsley if they can’t recall what they taste like. However choosing between French Fries and Sweet Potato Fries would be more universally accessible (at least in the West).
Where Conjoint Analysis might not be appropriate is when the sum is more than its parts. For example a painting like the Mona Lisa couldn’t easily be explained by its attributes. It can also run into trouble when some attributes pair much better with others, creating a local minimum. For example Dauphinoise Potatoes and Steak go well together, Burgers and Fries do too, but run a Conjoint analysis and combine the highest coefficients, and you might end up with a monstrosity. Burger Dauphinoise. You have to be careful about segmentation and grouping in these scenarios. For example survey groups of users that have similar preferences, or remove pairings that make no sense from the analysis.
There’s no reason conjoint analysis has to be limited to product attributes: its works on most things consumers choose between. If you're building houses, you could determine the part-worth utility of installing a swimming pool vs. the total price vs. the square footage vs. access to public transport, in order to prioritize your investment budget. As a marketer, the ideal situation to use Conjoint Analysis is before a new product is launched, to narrow down with data the best combination of features that consumers prefer. Post launch it can also be used to break down your ad creative into its component parts, or memes, so you can test different combinations to identify what’s driving clicks and conversions.
Name | Link | Type |
---|---|---|
Conjoint analysis | Reference | |
Conjoint Preference Share Simulator | Blog | |
What everyone gets wrong about this famous Steve Jobs quote, according to Lyft's design boss | Quote | |
What is Conjoint Analysis? | Book |