1 Introduction

Consumer decision-making has changed with rapid technological advances including mobile technology. Physical retail stores have become merely one of many sources of product information for consumer decision making, along with a variety of online sources such as ecommerce sites, manufacturer sites, and online social media where user-generated product information (e.g., expert and customer reviews and ratings) is shared. With the abundance of information always available within a few clicks/taps, consumers no longer rush to make purchase decisions while they are in the store. The overload of product information makes it hard for consumers to acquire and process it fully within the store, motivating them to delay decisions until they have had the opportunity to review and compare choice alternatives online at a location and time of convenience to them. This trend naturally has led to the ever-increasing ecommerce sales [1]. In-store retailers are now compelled to offer consumers reasons to shop in the store instead of other channels, suggesting an acute need to reinvent their services to enhance customers’ abilities to make decisions while in stores. In-store mobile decision aids (MoDA), which facilitate the acquisition and processing of product information and purchase decision making in the store while being able to physically examine the products, may address this need and provide in-store retailers with a competitive advantage over online retailers. In this paper, based on Chattaraman, Kwon, Eugene, and Gilbert’s [2] Human-Elaboration-Object-Construal (HEOC) Contingency Model, we propose design principles for modeling conversational flows between a user and a language-based in-store MoDA which functions as a virtual sales associate and provides context-aware decision assistance in the way that caters to individual consumers’ decision goals (e.g., product attributes or benefits sought) and constraints (e.g., time, product knowledge, cognitive resources).

2 Modeling In-Store MoDA Conversational Flows

The HEOC Contingency Model [3] postulates that an intelligent decision support system can predict a user’s decision-making strategy based on the user’s levels of decision elaboration (whether the user is likely to exert high or low effort in deliberating on the decision) and construal (whether the decision deliberation focuses on alternatives or attributes) identified from the user’s decision preference input. Brand names and model names are examples of alternatives, whereas product features and functionality represent attributes. Specifically, the HEOC contingency model delineates the prediction of four common consumer decision-making strategies: lexicographic or LEX (low elaboration, attribute focus), satisficing or SAT (low elaboration, alternative focus), elimination by aspects or EBA (high elaboration, attribute focus), and weighted adding or WAD (high elaboration, alternative focus) [3]. Based on this model, we conceptualize a language-based in-store MoDA which

  1. 1.

    performs an on-going assessment and verification of the presence/absence, number, type, and specificity of the user’s decision preference from his or her spoken input, with a goal of identifying the user’s levels of decision elaboration (high vs. low) and construal (attribute vs. alternative focus),

  2. 2.

    predicts the user’s adoption of and shifts across the four decision making strategies (i.e., LEX, SAT, EBA, WAD) based on the identified decision elaboration and construal levels,

  3. 3.

    identifies MoDA roles (e.g., decision preference prioritization aid, decision choice aid) and tasks that are relevant to the predicted decision-making strategy (see Table 1),

    Table 1. In-store mobile decision aids (MoDA) conversation flow design principles by user decision-making strategy: roles, tasks, and intelligence
    Full size table
  4. 4.

    generates or accesses strategy- and task-relevant MoDA intelligence (see Table 1 for MoDA intelligence needed for each user strategy and task), and

  5. 5.

    utters strategy-, task-, and intelligence-relevant speech to naturally support the user’s decision-making strategy (see Fig. 1 for the model and example of MoDA conversational flows for LEX decision makers).

    Fig. 1.
    figure 1

    Model and example of MoDA conversational flows for LEX decision makers. Note: dashed arrows are flows for User 1, and solid arrows are for User 2.

    Full size image

For example, through a conversation with a user, MoDA may find that the user is lacking product domain knowledge and unable to articulate decision preferences (see User 1 in Fig. 1), in which case MoDA initially predicts the user as a low elaborator who could be helped by learning about product attributes (i.e., a LEX decision maker). With this prediction, MoDA would now play a role as a decision preference prioritization aid (DPP_A) by recommending an important attribute (see Fig. 1 and DPP_A Task #2 in Table 1) and describing this attribute to the user (see Fig. 1 and DPP_A Task #3 in Table 1), and then shifting to a role as a decision choice aid (DC_A) by proposing an alternative that performs best on the recommended attribute (see Fig. 1 and DC_A Task #1 in Table 1). On the other hand, if the conversation with the user reveals that the user has a clearly preferred product attribute (see User 2 in Fig. 1), although MoDA may still classify this user as a LEX decision maker, it can skip the DPP_A role and directly engage in a DC_A role by proposing the best-performing alternative on the user-voiced attribute. In other scenarios (e.g., when user input reveals his or her preference for certain alternatives such as brands or models [i.e., SAT or WAD] and/or indicates the user’s knowledge or motivation to process elaborately many types of attributes [i.e., EBA or WAD]), their respective roles, tasks, and types of intelligence implemented by MoDA will vary as outlined in Table 1.

3 Conclusion

Most intelligent agents in consumer environments such as ecommerce sites have served merely as navigational/procedural aids (e.g., Alaska Airlines’ Jen). Previous literature on recommendation agents has assumed that users have well-defined decision preferences [4, 5], and hence has lacked in developing the aids for preference formation and prioritization. Previous work has also assumed that users are able and motivated to engage in an elaborative and rational decision-making strategy for an “accurate” decision [6,7,8,9]. However, consumer decision making tends to be constructive (i.e., consumers may form preferences as they learn about options) and is characterized by shifts across multiple decision-making strategies. Further, decision aid literature is scant on decision aids for the spoken language-based interface, which is the most natural mode of interaction between human agents and consumers in stores (cf., [10, 11]). The in-store MoDA conversational flow design principles proposed in this paper, based on the HEOC Contingency Model, articulate the specific roles, tasks, and types of intelligence (ability and knowledge) to be implemented in designing language-based intelligent agents that understand the consumer’s use of and shifts across four common decision-making strategies during an in-store shopping process. The proposed approach contributes to advancing the intelligent agent literature by enlightening the user intent aspect of natural language understanding (NLU), which is a key area for current and future artificial intelligence research.