In order to predict, you need a really robust data set. Consider the customer data you’re able to store in a CDP. You might know how a customer interacts across different devices and channels, their demographic information and what they’ve bought in the past, but this is not robust data - it represents a tiny fraction of a percent of the data points that would be available on any given person and how they spend their time online.
Think about how long you spend on a website and how much information you willingly give. Does this fraction of data shared capture the essence of who you are or what you want? Often not. For this reason, customers cannot be commoditized. Every single person is completely different. Their context and their intent is constantly changing. They might have bought an item for themselves from a site a few weeks ago, but today they could be shopping for their great aunt, or their brother’s newborn baby.
It’s also important to understand that customers are not the same as their demographic. What do we mean by this?
Consider Ozzy Osbourne and Prince Charles - two people with the same demographic data on paper. Same age, both are wealthy men, and both live in castles in England. And yet, you’d imagine their shopping carts look very different! With this in mind, trying to serve Ozzy predictions based on Prince Charles’ purchasing behavior doesn't make any sense. On the other hand, if you knew the items that Ozzy had recently looked at, and you knew what else customers who viewed those same items went on to click on or buy, you could serve Ozzy much better predictions based on this knowledge.
For exactly this reason, it is product data, not customer data, that is most effective for prediction in ecommerce.
Product data is valuable because you have:
- 100% of your item data.
- Every single item interaction.
- Every item purchased.
- Every search.
- Every item image.
- All of your product description data.
When you are able to understand the relationships between items (which items are similar to one another, which items go well together, which items are complementary to one another), then you can make predictions based on the in-moment intent of a customer, rather than demographic data or information about what a customer bought six months ago.
Advanced personalization platforms like Particular Audience use machine learning tech including computer vision to serve visually similar recommendations, as well as collaborative filtering tech to harness the wisdom of the crowd. By learning from the shopping journeys of hundreds of thousands customers, Particular Audience can better predict what customers want - if 100,000 people have followed a similar browsing journey, based on what they did next collaborative filtering can shortcut a customer’s path to the same outcome.