article
T1A Optimizes Discount Rates for Major Online Retailer Using AI
One of the biggest East European retailer and T1A have finished the pilot project determining the maximum discount rate for personal offers aimed to reactivate the customers prone to churn. The test group has shown a 37 percent increase in sales revenue and an 11 percent surplus in absolute margin compared to the control group as the result of the project.
Background
Retailer works hard on customer retention and often follows the approach when the customers prone to churn receive the X promocode. This strategy has been developed through the series of A/B tests of various offers. Thus, the promocode with X discount has proved to be the most efficient offer. The conversion rate of the promocode was two times higher than other options. However, the problem remained, and the company had reasonable doubts whether this approach could be considered as the optimal solution to all customers.

Solution
The business objective of the project was to raise efficiency of advertising campaigns among test group of customers. It was required to ensure the following results:
  1. Increase profitability of marketing communications for this segment.
  2. Expand the share of returned customers.
The retailer’s team and experts from T1A have formulated the following hypothesis: “the X promocode, indeed, fits the majority of the customers, but there are also buyers for whom promocodes Y, W or Z fit much better. It is required to learn to determine discount rates for individual customers''. The taskforce has developed the pool of models forecasting customers’ sensitivity to promocodes.

All the models are based on the following attributes:
  • Character of the latest orders, for instance, change of order content or regularity of orders.
  • Average bill and the share of the purchases split by price segments.
  • Interest in special offers and discounts.
  • Activity on the web site and in the mobile app.
  • Geography of orders.

Thus, the customers did not receive one and the same offer, but they started to get one of the four offers that was the best-fit for each individual customer based on the customer’s order history. For example, for the customers, who had made big purchases when they were active, it was preferable to provide a promocode with a big discount on a big shopping bill.

Personalization of discounts had positive impact on effectiveness of the marketing campaigns. The test group demonstrated 37 percent surplus to sales revenue and 11 percent growth in aggregate absolute margin compared to the control group as the result of the advertising campaigns.

Conclusions
The current T1A’s case is the bright example of transition from the state “one offer for the entire macro segment of customers” to the next level that implies “personalized offering for each and every customer”. For retailers it is vital to work with customer attrition because of the severe competition on the one hand, and significant positive effect from every returned customer on the other hand. The taskforce has developed quite a flexible mechanism to determine optimal discount rate. Moreover, it is quite simple to exclude or add new offers to the model.

The partnering teams plan to test additional promocodes to be included into the pool of models in the future as well as to apply the successful logic to other customer segments.