Sequential consumer choice as multi-cued retrieval

Abstract

Whether adding songs to a playlist or groceries during an online shop, how do we decide what to choose next? We develop a model that predicts such open-ended, sequential choices using a process of cued retrieval from long-term memory. Using the past choice to cue subsequent retrievals, this model predicts the sequential purchases and response times of nearly 5 million grocery purchases made by more than 100,000 online shoppers. Products can be associated in different ways, such as by their episodic association or semantic overlap, and we find that consumers query multiple forms of associative knowledge when retrieving options. Attending to certain knowledge sources, as estimated by our model, predicts important retrieval errors, such as the propensity to forget or add unwanted products. Our results demonstrate how basic memory retrieval mechanisms shape choices in real-world, goal-directed tasks.

Fig. 1.. Deciding what to choose next when shopping for groceries online depends on cued retrieval from multiple knowledge sources.

(A) We used 4.3 million unordered, in-store receipts to build representations of episodic, semantic, and hierarchical knowledge. (B) To model retrieval, we collected data from 135,000 shoppers as they sequentially searched for products on the website of one of the United Kingdom’s largest supermarket retailers. (C) Prior choices predict future ones, by virtue of their similarity according to different representational formats. Once an item is added to their basket, shoppers use this to cue matches from long-term memory. The stronger the match with this cue, the higher the probability an item will be retrieved (this may be attenuated by increased attention toward a particular representation). Retrieved items are checked against one’s internal goals. If the retrieval is goal relevant, then the shopper adds an appropriate item from the website and uses that item to cue associations. If not, then a new option is retrieved and checked for goal relevance until one is accepted. Similar heuristic strategies have been used in models of option generation for single choices (18, 19). Once all goals are satisfied, the user checks out. Note that the goal-checking process is not modeled here.

Fig. 2.. Consecutive purchases tend to be close episodic, semantic, and hierarchical relations.

(A) Choices are predicted by their similarity with the prior choice across each representation. Histograms show that the similarity between consecutive purchases (averaged for each visit) was higher compared to when the order of purchases was randomly permuted (with 95% CIs). (B) Sequential retrieval is similar to a ripple in semantic memory. Mean episodic similarity (with 95% CIs) between the current product and those purchased recently is higher compared with products purchased later. (C) Visitors slowed as they approached the end of their shopping trip. Mean response times (with 95% CIs) as a function of timestep quantile (small, 10 to 30 items; medium, 31 to 49 items; large, 50+ items). (D) Consumers make more between-category transitions (i.e., taxonomy level five) toward the end of their visit. Stacked density plots denoting the proportion of switches according to each level of the taxonomy as a function of the relative timestep. (E) Transitions between product groups at the fourth level of the hierarchy clustered into intuitive higher-order groupings that appear similar to those in the product taxonomy, suggesting that the taxonomy closely resembles how shoppers represent products during sequential choice. The Lift-1 of each transition is depicted in purple, with values less than 0 shown in gray. Boxes represent clusters identified by the optimal spectral clustering solution (more information in sections S1.7 and S2.6). ROI, Republic of Ireland; HNW, Health & Wellness.

Fig. 3.. Mean number of forgotten items (with 95% CIs) for each model attention weight (β).

Results show that relying on episodic or hierarchical knowledge predicted fewer forgotten items, whereas attending to semantic knowledge predicted more forgotten items, as measured by the use of a recommender system displayed before checkout.