How AI manages CPG’s couponing budgets more efficiently
Retailers, brands and couponing providers share a common challenge: how to best manage a limited promotion budget and prevent overshooting it when offering coupons to consumers? While the cost of distribution can be easily calculated, the cost of discounting (aka redemption) is difficult to predict.
During execution of our SO1® Smart Recommendation solution at a major US retail client, we could prove an interesting side effect that I want to share today. It illustrates nicely that our AI Engine can also handle (budget) constraints and deal with ad-hoc imposed changes much better than a system that at least partially relies on human interaction.
Originally, redemptions from two groups of weekly email recipient’s were monitored, where one group was targeted with our AI-based offer composition while the other served as a control group with users being treated by a rule-based system. Both systems could choose their recommendations at will from an identical pool of offers. These offers were paid for by CPG brands, based on pre-defined budgets.
Business intervention limits campaign distribution
Approximately six weeks after launch, the retailer realized that both systems had distributed offers in a way that had partially exhausted the CPG promotion budgets which were negotiated with the brands on a one year base. Reacting to this, the retailer removed those campaigns from the available pool that had already exceeded the budget. On other campaigns, the retailer introduced distribution limits, i.e. a maximum number on how often an individual campaign could be recommended and redeemed.
*limited distribution ~ Offer can be distributed to less than 5% of Email recipients
Fig. 1: Left: Redemption rate and relative performance advantage over time for SO1 Offers (blue) compared to a rule-based system (red). SO1 offers outperformed rule-based recommendations. But as constraints on the campaign pool are introduced (number of offers & distribution limits), the relative performance advantage of AI offers increased significantly.
Right: The number of available campaigns decreased in week 13 by 1/3, where in addition more than 90% of the offers came with strict limits.
The effect on performance was striking. While the absolute offer redemption rate, as expected, decreased in both groups (due to less variety and removal of well-performing campaigns), the negative impact on the rule-based system was much stronger and the relative advantage of the AI Offers doubled from originally 57% to a 132% higher conversion rate.
Why does performance advantage increase?
Well, the SO1 Engine automatically understands product properties and consumer preferences on a very deep and granular level, i.e. for every single product and for every single customer. Consequently, the targeting is much more precise and efficient – as shown in the following example of recommending eggs: While the SO1 AI recommended the campaign to less than half as many households (2.8%) compared to the rule-based system (8.6%), the targeted households redeemed the SO1 offer at a much higher rate – demonstrating the superior targeting performance of the SO1 AI.
Fig. 2: Exemplary performance comparison of a popular campaign for Eggs & Eggs Substitutes, taken from week 4 (see Fig.1)
This example also shows the high flexibility of the SO1 AI in dealing with an unforeseen business constraint. With no effort and adaption time, it considered the distribution and budget limits and still outperformed the rule-based system – thus preventing the retailer from overshooting the pre-defined brand budgets.
The SO1 AI is precisely built to manage campaigns automatically according to given constraints and business goals (e.g. maximize redemptions, minimize discounts, empty stock etc…) and free the retailer’s staff from time-consuming evaluation, planning, and execution tasks.