In Switzerland, there has been broad resistance when local grocery retailers, such as Migros, decided to target customers more openly with personalized price discounts. Although the same retailers have already used personalized price discounts before, the fear of the transparent customer and personalized prices (not discounts) predominated the public opinion. Yet, elsewhere personalized price discounts belong to the retailer’s established toolkit. Despite the obvious disputability of the topic, investigating how the timing of a personalized price discount (not its height) influences the relevance of the promotional offer and its redemption is of high importance for marketing practice. First tests have already shown that by incorporating known timing information when providing a personalized discount, retailers can improve their ability to correctly predict redemptions by up to 6%. These improvements are expected to further increase when more sophisticated modelling techniques are applied. This enables retailers to provide discounts which are more likely to be redeemed and emphasizes the importance of the timing of a recommendation for its success.
Whether a customer redeems a price discount relates to her inter-purchase time. Consequently, a customer is most likely to participate in a promotion when the promotion is relevant to her and she is in actual need of the promoted product. Intuitively, purchase (and consumption) cycles are highly category-specific. Most customers, for instance, are likely to buy (and consume) bread more frequently than coffee, and coffee more frequently than laundry detergent. If a customer has bought bread five days ago, she might want to purchase a fresh loaf today as she just finished the old one. A price discount would thus come at hand. If, however, she just bought bread yesterday and visits the grocery store again today, a recommendation and price discount for bread might be less useful from her perspective. Incorporating inter-purchase times into the process of recommendation generation should encourage recommendations which are more relevant and more likely to be redeemed. Yet, it is important to note that a retailer cannot observe purchases that are made at competing retailers. Consequently, the observed inter-purchase times are not the true category inter-purchase times. This important detail renders the search for the optimal point in time for a recommendation more complicated than it might seem at first glance.
To find the optimal point in time for a recommendation, a closer look is taken at a provider of an in-store recommender system at an offline grocery retailer. The in-store recommender system provides a customer with personalized recommendations when she enters the retail store. The recommendations are made based on the past transaction history but do not include information on the timing of past purchases or customer demographics. After scanning the loyalty card, a list of tailored price promotions is printed. At the check-out, the price discounts are automatically redeemed when the loyalty card is scanned again.
Analyzing the data of the recommender system shows that the optimal point in time for a recommendation is highly category-specific. For instance, for some fast-moving consumer goods, such as bread, curd, or milk, redemption rates of the personalized recommendations are best described by an inverse u-shape. This implies that there is an optimal point in time for a recommendation in a product category. Consequently, redemption rates are lower when the recommendation is provided too early or too late. One might be induced to think that the optimal point in time coincides with the average observed inter-purchase time, but this is not the case. This can be explained by the fact that the observed inter-purchase time is not the true category inter-purchase time—due to unobserved purchases at other retailers. By just incorporating known information on the optimal point in time for a recommendation and the time since the last category purchase, the retailer can improve his ability to correctly predict redemptions by up to 6%. This highlights the potential for improvement if so far unknown information on the true category inter-purchase time was leveraged to improve the timing and the success of recommendations.
What is the advantage of providing customers with personalized recommendations in the form of price discounts? From a customer perspective, optimizing the timing of a recommendation improves its relevance. Customers consequently get recommendations for products they are actually interested in and are more likely to redeem the recommendations. From a retailer perspective, personalized recommendations can be used to promote customer loyalty. First, customers who want to benefit from these special targeted offers need to come to the store to receive them. Second, personalized recommendations, particularly when they come at the right time, can be used as a tool to shift purchases from competitors to the focal retailer. As most customers shop at several retailers, their loyalty to the focal retailer might be high in some product categories but low in others. Providing recommendations at the right time might thus increase the customer’s share of wallet in categories that she frequently purchases at different retail chains.
Given that retailers can only observe purchases made at their own stores, optimizing the timing of personalized recommendations remains problematic. Purchases at competing retailers are unknown and so are the true category inter-purchase times and the true category share of wallets. Research to infer these unknowns from observations made at a single retailer has great practical relevance and can be applied to further improve the timing of personalized recommendations. Finally, even though the height of personalized price discounts seems to be at the center of public attention, their timing is just as important for customers and retailers.
Based on the following paper:
Vuckovac, D., Wamsler, J., Ilic, A. and Natter, M. (2016, September), “Getting the Timing Right: Leveraging Category Inter-purchase Times to Improve Recommender Systems”, Proceedings of the 10th ACM Conference on Recommender Systems, 277-280, ACM.
Further information about the author: Julia Wamsler