“Inspired by the human brain and its neural network when understanding what products to recommend,..”
Beyond Retail met Oliver Edholm, AI researcher and co-founder of the new product recommendation platform depict.ai, a platform challenging Nosto and other big competitors within product recommendations and e-commerce personalization. The meeting made us deep-dive into the logic behind product recommendations, and what separates depicit.ai from previous solutions.
Historically, players within product recommendations have all used different versions of the same algorithm when creating their product recommendations. The logic has been to look at what “others also bought”, referred to in research terms as ”collaborative filtering”.
However, Depicit.ai (and we are sure others will challenge them) use a new algorithm and logic. They argue that the old logic “others also bought” only works for the really big players with enough data to “fuel” the algorithm. Most e-commerce players simply do not have enough data to use the old logic, and something called the “cold-start problem” appears when you don’t have enough data to work with.
The problem with the old “others also bought” logic
- Few e-commerce businesses have enough data to make it work (very few are Alibaba)
- Not optimal if you are selling a lot of seasonal products (as new products don’t get recommended)
- A loop is created where only popular products get recommended (ending up with very generic recommendations)
- Time consuming to gather enough data
- Does not consider any product information
The new logic inspired by the human brain
Depict.ai have found a way to solve the so-called “cold-start problem”. Using technology that was first available four years ago, they explain that they can understand the products through looking at them more similar to how a human being would. Rather than just looking at the transactional data of the products, they explain that they are inspired by the human brain and its neural network when understanding what products to recommend, looking at additional data points enabling several product recommendation types.
Additional data points
- Cookie data
- Trends & Season
Product recommendation types
- Similar products (taking in consideration visual attributes and text analysis)
- Up-sell (considering price sensitivity and popularity)
- Cross-sell (considering transactional and sessions data)
- Dynamic (considering real-time behavior and consumer intent)
It was clear to us that we were introduced to a new generation of AI for product recommendations. Regardless of the service provider your brand has chosen for product recommendations, we believe it is inspiring to see what additional data points are relevant and what different product recommendation types there are to consider.
At Beyond Retail we are independent from e-commerce service providers, but we also stay in the loop when something new comes along to challenge the status quo. We look for new possibilities for our clients, and recommend things that were previously not possible. Please reach out to us if you would like to discuss what services could optimize your competitive edge.