Applying AI to individualize loyalty card promotions in grocery retail
The retailer: One of the top 3 full assortment grocers in Germany
The retailer is one of the top 3 German grocery retailers which has over 20 billion Euro yearly revenue and more than 2.500 stores, being on average 1.000 – 1.500 sqm in size. The retailer has a well-established loyalty card system, which launched about 10 years ago and varies in customer acceptance throughout the retailer’s regions.
The challenges: Low acceptance of mass promotions and heuristically targeted coupons
The German grocery retail market is one of the toughest and home to worldwide competitive discounters such as Aldi and Lidl. Therefore, the retailer was confronted with huge price pressure, and low margins (market average is 0.4% EBITDA). Traditionally, the retailer was making use of mass promotions and targeted couponing based on socio-demographics and simple heuristics, but wanted to focus on a more personalized strategy. Said mass promotion activities saw low redemptions and caused the retailer to constantly lose margin on wrongly targeted consumers and promotions that frequently replaced regular sales. The retailer’s bad experience was in line with the results of a Nielsen study, that showed that globally 59% of mass promotions do not break even.
The proposed solution: Individual promotions powered by Artificial Intelligence
SO1 helped this retailer return their promotions to a source of incremental value, by providing personalized offers to each individual loyalty card customer based on SO1’s Artificial Intelligence.
SO1 has created a powerful AI for retail which is capable of understanding consumer preferences, predicting a consumer’s willingness-to-pay for each product in a store and therefore personalizing promotions for users in real-time and across communication channels. The SO1 Engine sources the entire portfolio of the retailer and automatically selects the right products for each individual consumer and adjusts discounts such that revenue, profit, or consumer satisfaction are maximized. These features are commonly known as Targeting, Recommendation and Price/Discount Optimization Engines.
For this retailer, the pilot project was set up to last 8 weeks and was restricted to 60 stores within the area of Berlin, Germany. The pilot was preceded by a joint integration period of 12 weeks. This integration period included the transmission of historical basket data to train SO1’s algorithms on (a) the specific assortment per store and (b) the specific customer behavior at this retailer. This procedure guaranteed best-in-class results starting with the first distributed offers.
Loyalty card customers received individual offer feeds containing 5 relevant products and individually adjusted discounts. Discounts were granted in the form of loyalty points. The customer just needed to scan his loyalty card at a Kiosk system positioned in the entrance area of the selected stores and his or her individual feeds got printed (check-in couponing). Offers were valid for the day of printing only. Deal-stacking was avoided thanks to an automated matching of offered coupons with the weekly’s circular offers.
Test setup: Retail experts versus SO1’s AI
In order to measure the success of AI-based individual promotions, the pilot project was accompanied by an A/B test. For this test, the retailer’s sales and marketing experts defined a pool of ~300 top promotions (e.g. Coca-Cola, different flavors, maximum 35% discount) to choose from. Users were randomly split into disjoint cohorts by their loyalty card ID; 80% of the users received SO1’s AI-based individual offers – i.e. SO1 distributed selected offers from the pool only to certain users for whom they were relevant and adjusted the discounts according to the users willingness-to-pay; the remaining 20% of the users received the offers composed by the retailer’s experts – these offers from the pool were randomly selected and distributed, while the discounts remained at the pre-defined maximum. The retailer then compared both systems in terms of redemption rates and average discounts applied.
The results: AI significantly outsmarts retail experts
The performance difference resulting from this A/B test was remarkable. Not only were AI-selected campaigns on average nine times more likely to be redeemed, the overall discount was only one third of the discount given in the expert based system.
Figure 1: Redemption Rate and Discount of individual offers based on Artificial Intelligence (SO1, blue) vs. promotions selected by the Retailer’s expert team (grey). Setup: Offers distributed via printouts with the identical layout at Point-of-Sale kiosk located at the store entrance (check-in). Source: Confirmed Retailer Data
Artificial Intelligence thus significantly increased the efficiency (return-on-invest) of this retailer’s price promotions by triggering successful conversions (increased sales) and at the same time saved 60% of the budget spent on discounts – directly resulting in increased margins.
SO1’s artificial intelligence is built to deliver financial impact for grocery retail. It can be applied in all fast moving consumer goods environments like grocery, drugstore, OTC, pet food and the likes and works for brick and mortar as well as online retail. SO1 supports various use cases and communication channels towards consumers (e.g. ranking existing leaflet/circular offers by customer preferences without altering discount levels) which means it is flexible in the extent of technical integration that is required to get started. It integrates easily into existing systems, tools and processes and adapts to different business goals to derive maximum efficiency.