- Customer analytics
- Recommendation system
- Clustering
- Machine learning
- Online retail
Key challenges
An online retailer in the automotive industry used to show its customers its entire product range in an unstructured way. As a result, his customers often felt overwhelmed by his large product range and did not want to search the entire store on their own. This resulted in a low probability of additional sales and a poorer customer experience.
Our approach
Based on historical sales data the experts of Positive Thinking Company created a customer and product segmentation.
First, customer profiles were generated that describe the historical buying behaviour including attributes like number of purchases, average market basket value or the share of different brands and product groups.
Next, a data science algorithm was run to cluster customers with a certain minimum number of purchases. The method produced particularly good separability of customers based on brand and product group. With this newly gleaned information, the whole range of products on the platform was reorganized and grouped according to brand on the first level, and according to product group on the second level.
The final segmentation produced around 30 well-distinguishable product clusters, for which those customers were then proportionately selected who made the most purchases in a cluster, based on their profile.
Benefits
The model developed enables the company to offer a product to those customers who show a very high preference for it and, at the same time, to hide products that customers won’t be interested in. Because cluster affiliations are recalculated every month, shifts in customer preferences are uncovered, so the model is always up-to-date.
As a result, potential customers were addressed in a more targeted way and their user experience was improved, because they only received relevant information.
