We’ve been blogging about a Canadian grocer that is changing the game with its analytics-driven loyalty program. For its +9 million loyalty members, personalized offers are so relevant, it’s almost like having an old-fashioned relationship with the corner grocer. In this new-fashioned relationship, however, members can redeem their offers at more than a thousand stores of various types, in many locations across Canada.
What’s this grocer’s secret to success? Here’s a peek inside the analytics.
Understanding customers as more than strings of transactions
Delivering a personally unique set of relevant offers every week on such massive scale is a computationally demanding, mathematically intense undertaking. The heavy lifting is performed by over 4,000 time-to-event (TTE) predictive models, generated and updated every four months by a FICO-built analytic “factory.”
Each TTE model predicts the propensity of a customer to purchase a particular product — say, a specific brand of laundry detergent — within a specific timeframe. Together these models output over 90 million offers per week — metrics that tell the grocer the likelihood, for every loyalty member, of buying each of some 60,000 products in the next seven days.
Analytics also help the grocer understand customer sensitivities to loyalty rewards. It can, therefore, focus incentives where they are most likely to not only cultivate loyalty but also produce profitable changes in customer behavior.
Identifying the best offers from tens of millions
Each week, TTE models generate 90 million offer recommendations for loyalty members. Optimization is used to select a maximum of 20 best offers per customer that maximize relevancy and fuel profit. The optimization engine in FICO® Analytic Offer Manager solves for this goal, while adhering to the grocer’s constraints that include concentrating a significant amount of incentive investment on its best customers.
Learning what causes customers to buy
To continue bringing loyalty members relevant, profitable offers each week — and get better and better at doing it — the grocer needs a constant stream of data that reveals cause and effect. Did the 1,000-point offer cause Carmen to buy a rotisserie chicken this week, or did the offer just correlate with a purchase she would have made anyway?
To know for sure, the grocer would need to observe what that customer does when presented with an offer compared to when not presented with that offer. Since it’s impossible to observe two opposite conditions on the same customer simultaneously, the next best method is to conduct controlled “test and learn” experiments among customers with similar purchase probabilities, as shown in the graphic below.
Download our white paper Loyalty Is Rocket Science for a Major Canadian Grocer to learn more about the analytics behind this client’s success.