Analysis of customer purchase data for an international FMCG company

Stage 1 – The mechanics of the loyalty program are fairly straightforward:

1. Advertisements are placed in stores, indicating that receipts with the brand’s products can be registered on the loyalty program’s website.
2. Customers create personal accounts on the website and upload their receipts.
3. Our integrated CV module reads the QR code on the receipt to extract purchase information and validate whether the purchase meets all campaign conditions.
4. The system automatically identifies the brand’s products and accrues points that customers can later spend – either exchanging for discounts or gifts from the brand.

Additionally, the website’s administrative panel gathers comprehensive statistics about all customers registered in the program and their activities. This data aids in evaluating the success of promotions.

After a year and a half since launching the loyalty program and accumulating 300,000 receipts, a valid question arose: how could we leverage the gathered data and what more could we learn about our customers? By analyzing their purchases, it becomes possible to gain deeper insights into the customer profile and identify which complementary products lead to the purchase of the brand’s items.

Stage 2 – At the beginning of 2021, we initiated the development of a platform for receipt analysis. We needed to establish a system for regular data analysis of purchases, enabling us to track dynamics. Now, with the existing 300,000 receipts, we had to derive not only information about the brand’s products but also answer three additional questions:

1. Which other brands are often purchased alongside the client’s brand?
2. Specifically, which products are they?
3. Do companion products change as the receipt value increases?

We encountered the challenge that the data quality was poor. As a reminder, we were collecting receipts from various stores within the extensive network, each of which operated as an independent entity. The same item could be entered into the system differently or different items could be entered in the same way, and the costs could also vary.

In order to structure the data, we developed our own parser. It consists of a vast array of rules that structure all the information and bring it into a unified format. It learned to extract all the necessary details about the items from the receipt, details that were crucial to understand, such as product lines, categories, and quantities. This allowed us to differentiate all the brand’s SKUs, but this was just one stream of information.

The second stream included all the other items on the receipt, which we also needed to recognize. Since we didn’t require such detailed information about these items, we integrated receipt recognition from the Bank, and that proved to be sufficient.

To visualize the acquired data, we developed an interactive dashboard using Microsoft Power BI, presenting different perspectives, and setting up monthly automated data refreshes.

With the information we’ve gathered, we can now compile comprehensive analytics regarding product adjacency. Based on this, the brand can:

• Conduct RFM analysis of customers and prepare personalized offers for buyers.
• Plan joint promotional campaigns with other manufacturers.
• Launch effective targeted advertising.

Furthermore, the loyalty program has integrated with the other marketplace, which means we can now obtain even more insights into customer preferences.