Developments in machine learning and big data allow firms to fully personalize and target their marketing mix. However, data and privacy regulations, such as those in the European Union (GDPR), incorporate a “right to explanation”, which is fulfilled when targeting policies are comprehensible to customers. This paper provides a framework for firms to navigate right-to-explanation laws. First, I introduce a new method called Policy DNN, which combines policy learning and deep neural networks, to form a profit-maximizing black box benchmark and provide theoretical guarantees on its performance. In contrast to prior approaches that use a two-step method of estimating treatment effects before assigning individuals their treatment group, Policy DNN directly estimates treatment assignment, which improves efficiency. Second, I construct a class of comprehensible targeting policies that is represented by a sentence. Third, I show how to optimize over this class of policies to find the profit-maximizing comprehensible policy. I demonstrate that it is optimal to estimate the comprehensible policy directly from the data, rather than projecting down the black box policy into a comprehensible policy. Finally, I apply my framework empirically in the context of price promotions for a durable goods retailer using data from a field experiment. I quantify the cost of explanation, which I define as the difference in expected profits between the optimal black box and comprehensible targeting policies. The comprehensible targeting policy reduces profits by 7% or 22 cents per customer when compared to the black box benchmark.
Advances in estimating heterogeneous treatment effects enable firms to personalize marketing mix elements and target individuals at an unmatched level of granularity, but feasibility constraints limit such personalization. In practice, firms choose which unique treatments to offer and which individuals to offer these treatments with the goal of maximizing profits: we call this the coarse personalization problem. We propose a two-step solution that makes segmentation and targeting decisions in concert. First, the firm personalizes by estimating conditional average treatment effects. Second, the firm discretizes by utilizing treatment effects to choose which unique treatments to offer and who to assign to these treatments. We show that a combination of available machine learning tools for estimating heterogeneous treatment effects and a novel application of optimal transport methods provides a viable and efficient solution. With data from a large-scale field experiment for promotions management, we find that our methodology outperforms extant approaches that segment on consumer characteristics or preferences and those that only search over a prespecified grid. Using our procedure, the firm recoups over 99.5% of its expected incremental profits under fully granular personalization while offering only five unique treatments. We conclude by discussing how coarse personalization arises in other domains.
We propose a framework for welfare analysis of nudges when consumers hold potentially biased beliefs. We estimate a structural model of the influence of nudges on consumer beliefs and purchase decisions in the context of a large food delivery platform. We identify the model by combining panel data on consumer purchasing behavior with variation from a randomized field experiment that nudged consumers to purchase the platform's subscription service. Consumers make a repeated monthly decision on whether to renew the subscription and they hold biased beliefs if they make repeated subscription mistakes from an ex-post perspective. Nudges have a substantial effect on on-platform consumption but they have an ambiguous effect on consumer welfare as only some consumers are made better off by purchasing the subscription. We show how policy makers can leverage these heterogeneous treatment effects and the structural model to optimally target nudges to maximize consumer or producer surplus.
We present a general framework to target customers using optimal targeting policies, and we document the profit differences from alternative estimates of the optimal targeting policies. Two foundations of the framework are conditional average treatment effects (CATEs) and off-policy evaluation using data with randomized targeting. This policy evaluation approach allows us to evaluate an arbitrary number of different targeting policies using only one randomized data set and thus provides large cost advantages over conducting a corresponding number of field experiments. We use different CATE estimation methods to construct and compare alternative targeting policies. Our particular focus is on the distinction between indirect and direct methods. The indirect methods predict the CATEs using a conditional expectation function estimated on outcome levels, whereas the direct methods specifically predict the treatment effects of targeting. We introduce a new direct estimation method called treatment effect projection (TEP). The TEP is a non-parametric CATE estimator that we regularize using a transformed outcome loss which, in expectation, is identical to a loss that we could construct if the individual treatment effects were observed. The empirical application is to a catalog mailing with a high-dimensional set of customer features. We document the profits of the estimated policies using data from two campaigns conducted one year apart, which allows us to assess the transportability of the predictions to a campaign implemented one year after collecting the training data. All estimates of the optimal targeting policies yield larger profits than uniform policies that target none or all customers. Further, there are significant profit differences across the methods, with the direct estimation methods yielding substantially larger economic value than the indirect methods.
Fišar, M., Greiner, B., Huber, C., Katok, E., Ozkes, A., and the Management Science Reproducibility Collaboration (2023)