Advanced AI-based customer segmentation

One of our clients – a global FMCG manufacturer – was looking for new ways to cluster and identify the purchasing behavior of consumers registered in a CRM database. The objective was to provide AI-based client segmentation and align the CRM data with existing personas. Assigning clients to particular personas was seen as a major factor in improving the efficiency of advertising campaigns, both in online and offline channels.

The challenge

Our client was already using behavioral segmentation based on personas, but it was struggling to link them with actual customers in the CRM database.

The solution

Data from all available touchpoints (web activity, 1-to-1 contacts, partner stores, emails and SMS) was used to train ML-based segmentation models. To keep the maximum amount of information about a client without allowing for a data leak, we developed two separate segmentations:

• Consumers with orders - using all available data, but trained and applied only to consumers with orders in the database,
• Consumers without orders - this segmentation was trained on all consumers but using only data prior to first purchase.

By applying different modeling methods to each of the above groups, we created a comprehensive approach to customer segmentation that brings together multiple sources of input. The resulting insights helped create a mapping between personas defined within the target group and CRM entries.

The effect

Based on that, we created new features to better capture consumer behaviour. In the final version, the machine learning model used about 200 features. Segmentation returned 13 separate segments.

What’s more, we linked identified segments with personas in existing behavioral segmentation. It was a crucial step, as behavioral personas are widely used by various functions, e.g. by marketing departments (to create dedicated campaigns for selected personas) as well as product departments (to define value propositions for different segments).

To that point, the client hadn’t been able to identify particular personas in his CRM base, so precise targeting was impossible. Our segmentation helped to answer the following questions:

• Which personas should I target?
• Do I have them in my consumer database? How do I find them?
• How can I reach them?

and lead to increased effectiveness of marketing activities.