Modeling New Car Customer-Salesperson Interaction For a Knowledge-Based System

Abraham D. Horowitz, General Motors Research Laboratories
J. Edward Russo, Cornell University
ABSTRACT - A model portraying a new car customer-salesperson interaction, derived from in-depth interviews with nine experienced salespeople, is presented. The model implies two parallel hierarchical structures: one for information processing, the other for the product. The present study follows the methodology of "knowledge engineering" in that model building is seen as an iterative process based on input from experts used to create and modify an expert system. The model is compared to selected studies of customer - salesperson interactions. It is suggested that testing the expert system constitutes a test of the model itself.
[ to cite ]:
Abraham D. Horowitz and J. Edward Russo (1989) ,"Modeling New Car Customer-Salesperson Interaction For a Knowledge-Based System", in NA - Advances in Consumer Research Volume 16, eds. Thomas K. Srull, Provo, UT : Association for Consumer Research, Pages: 392-398.

Advances in Consumer Research Volume 16, 1989      Pages 392-398


Abraham D. Horowitz, General Motors Research Laboratories

J. Edward Russo, Cornell University


A model portraying a new car customer-salesperson interaction, derived from in-depth interviews with nine experienced salespeople, is presented. The model implies two parallel hierarchical structures: one for information processing, the other for the product. The present study follows the methodology of "knowledge engineering" in that model building is seen as an iterative process based on input from experts used to create and modify an expert system. The model is compared to selected studies of customer - salesperson interactions. It is suggested that testing the expert system constitutes a test of the model itself.


Our immediate goal is to develop a conceptual model of a dialogue between a consumer and a salesperson in new automobile purchase situations. That model serves as the basis of a computer system that helps consumers select carlines, models, and optional features. We refer to this system as ESPRI (Expert System for Product Recommendation and Information). It should be able to execute some of the functions currently performed by salespeople, viz. matching an individual's needs and preferences to product options, accessing useful databases like technical specifications and availability of the chosen car, and offering audio/video information via new interactive technologies like videodiscs.

The present paper describes the derivation of a conceptual model that provides the frame on which ESPRI is being built. We compare this model to selected studies of consumer-salesperson interactions, and discuss how testing ESPRI constitutes a test of the model itself


Interviews with nine salespeople were conducted in four Pontiac car dealerships located in the Detroit metropolitan area. The sales manager of each dealership was asked to choose two salespeople to be interviewed. One was among the very best and the other average in terms of annual sales, though both had to have at least two years experience. (The sales manager of one dealership was also interviewed because of his expertise in training new salespeople). Salespeople were asked to describe in detail how they interact with customers, especially in terms of the dialogue that they maintain. In addition, they were asked to answer the following questions.

(1) What do customers know about the cars they are interested in at the outset of the interaction?

(2) What kind of product information do customers typically seek and what information do they usually not ask for?

(3) What information do salespeople request from customers?

(4) What information do salespeople voluntarily supply to customers?

(5) Under what conditions do salespeople recommend a specific automobile, and how is this recommendation communicated?

(6) Are different strategies used for different customers? If so, what are the different types of customers and how are they identified?

Each interview lasted from 45 to 60 minutes and was sound- recorded and transcribed for analysis. Responses were tabulated for each of the questions listed above, separately for superior and average salespeople.

The transcripts were analyzed informally with the goal of extracting a model of the customer-salesperson interaction that reflected a consensus of the nine interviews as-discussed in more detail below. We achieved this first by pruning details, sticking instead to the main flow of the dialogue which turned out to be surprisingly uniform across both superior and average salespeople. This first pass at a model was then shown to four of the original salespeople (at two randomly chosen dealerships). The model was used to simulate two typical customer consultations with ESPRI, and the salespeople were prompted for their reactions. While reacting positively to ESPRI's general approach, salespeople made specific suggestions for changes. Thus, the model presented in the following section was built in the spirit of "knowledge engineering," (see e.g., Harmon and King, 1985), in that model building is seen as an iterative process based on input from experts used to create and modify a computer software system.


Detailed interview analyses suggest that superior and average salespeople share a common view of the major components of what they do or should do in interacting with customers. (Variation in their sales performance is probably due to motivational and personality factors, i.e., the ability to execute a plan rather than having a plan, topics not addressed in this paper).

Analyses indicate that salespeople have a sequence of goals that can be grouped into seven stages:

I. Identify the customer's preferences by a few major screening attributes (e.g., bodystyle and car size).

II. Provide basic product information for the preferred cars and help customers request information and evaluate alternatives on self-selected major features.

III. Identify customer objective needs in terms of expected car use.

IV. Reconcile preferences and needs.

V. Let the customer make an informed choice among carlines.

VI. Evaluation and selection of a model: provide information, identify preferences, recommend, and let the customer select one model (of the chosen carline).

VII. Minor options: provide information, identify preferences and needs, recommend, and let the customer select options for the chosen car model.

This model of the customer-salesperson dialogue implies two parallel hierarchical structures: one for information processing, the other for the product. Information processing starts with elimination of alternatives based on a few major aspects. It continues with customer evaluation of the remaining alternatives, and ends with a product choice. The hierarchical product structure for automobiles is Carline-Model-Options. The two structures are nested such that information processing occurs separately for carline, model, and options choices.

Figure 1 describes the conceptual model in flow-chart form. It recognizes that the salesperson and the customer might return to a previous stage. These possibilities are denoted by dotted lines. Each of the seven stages is discussed next.

I. Identify Customer's Major Preferences

According to the interviewed salespeople, the first mentioned major preference attribute is bodystyle: station wagon, four-door, two-door, hatchback, sports, or convertible. Customers typically have strong feelings about their preferred bodystyle. Some customers will consider only a station wagon, while others want a sports car and would never consider a station wagon.

The second most often mentioned major attribute is carline name or car size. Customers are of two types: those who express preferences by naming one or two specific carlines, and those who specify their preferred car size or sizes. Customers are typically unfamiliar with the size terminology like compact, small, midsize, etc., and they use their own size terms which vary from one customer to another. Therefore salespeople present one or two cars, so as to match each customer's description. In addition, salespeople say that customers frequently consider two cars with the same bodystyle that differ only one level in size (e,g,, midsize and fullsize, or midsize and compact). They evaluate both more thoroughly and eventually select one. Customer preferences for bodystyle and carline (size) help the salesperson screen out all cars but one or two.

Price is typically not mentioned in Stage I either by customers nor by salespeople. The main exception is those cases where customers come to the dealership for the specific purpose of finding out the price of a specific car or for negotiating a price.

II. Provide Information

In this stage customers take the initiative in the dialogue, requesting information about car characteristics of special interest. Although there is wide variation among customers in the product information requested, according to our salespeople/experts the most frequently requested visual/verbal information is for: (1) exterior styling; (2) interior features, especially front seat design and instrument panel; (3) trunk size; (4) gas tank capacity; (S) the power-train, including front/rear wheel drive, manual/automatic transmission, and engine characteristics; (6) fuel economy; (7) exterior and interior colors; (8) new car price and trade-in value; and (9) comparisons with similar cars available from other car manufacturers.

Customers generally do not ask about car dimensions and do not raise questions about safety (e.g., brake types). Questions from customers about reliability, service, warranty, and maintenance costs usually occur after customers have made a tentative decision on the carline they wish to purchase (Stage V).

As indicated in Figure 1, salespeople provide information in two steps. First, they provide basic information addressed to all customers for one or two carlines. This information is an overview of the unique characteristics of the carline(s) with emphasis on visual cues of the exterior, interior, and under the hood, and other basic information, such as price. In the second step, salespeople adapt their presentation to specific customer information questions. For example, some customers are interested mainly in performance, while other are interested in exterior styling and/or economy aspects. Finally, customers react to the information provided by expressing their preferences among major options (e.g., standard engine vs. optional engine) or, if two carlines are considered, for one or the other based on exterior styling, price, roominess, etc. This process typically ends with an overall customer evaluation of and preference among alternatives with some guidance from the salesperson. Salespeople emphasize that they let the customer make their own "trade-offs" among alternatives.



In light of new information, the salesperson and/or the customer may move back to Stage I for consideration of a different carline.

III. Identify Customer Objective Product Needs

In this stage the salesperson makes the key decision about whether to identify needs. The salesperson will not identify needs in two cases. First, "sports" and "convertible" cars are typically bought for their unique sporty image rather than for functional utility. Second, certain cars, typically large and midsize cars, satisfy the needs of most drivers and/or families. If a customer's size preference is for small cars (say, smaller than midsize), the salesperson will proceed to identify the customer's specific needs. Customers who buy small cars may be later dissatisfied if the car does not accommodate family needs, especially if the car is used for vacation trips.

The salesperson prompts the customer for information on expected car use. If the intent is for family use, then he prompts for information on family size, expected long trips, and cargo carrying needs. The underlying rationale is that longer trips and larger families indicate the need for a larger car. Salespeople emphasize that the recommendation should not be based solely on family size since for a given family size, the longer the trip, the more cargo capacity and passenger space are needed to provide adequate comfort. Salespeople also point to the importance of identifying needs because customers often are not aware of the full product line.

IV. Reconcile Preferences and Needs

Once the salesperson establishes the customer's needs, the salesperson's role is to judge whether the preferred car(s) (P1, and, possibly, P2) meet those needs. If the gap is too large (the maximum tolerable gap size may vary among salespeople), the salesperson will explain why the recommended car (R) may better meet the customer's needs than P1 (and P2).

Salespeople say that they will not recommend a car that is "far apart" from what the customer wants; they strike a balance between preferences and needs.

V. Let Customer Select a Carline (Make an Informed Choice)

The previous stages in the model are dynamically interrelated in a process where the customer, using product information, might evaluate alternatives and eventually make a choice. The goal of the salesperson at this stage is to encourage the customer to make an informed choice. The previous stages have built up to this point.

VI. Model Evaluation and Selection: Provide Information, Identify Preferences, Recommend, and Let Customer Select One Model

Once the customer has focused on one carline, the customer - salesperson dialogue proceeds to the selection of a model (e.g., LE, STE). This may happen very early in the dialogue if the customer knows from the outset what car he/she wants. According to salespeople, less than 10% of customers are familiar with the various models available in a given carline. Therefore, an important role for the salesperson is to assist in model selection.

First, the salesperson provides information on what is unique for each model in the chosen carline, including price. Although this model-specific information may be redundant with Stage II, the emphasis now is on comparisons between car models. If the customer cannot make up his/her mind on a preferred model, the strategy used by salespeople is to identify the customer's "generic" preferences that may determine which model to recommend. Generic preferences refer to general attributes, such as luxury, roominess, and sportiness. Knowing customer's generic preferences helps the salesperson to recommend a model.

In addition, some specific features/options are available only in some models but not in others. For example, a carline may not offer bucket seats with a console or two-tone paint in all models.-Prompting the customers to express their preferences helps the salesperson to recommend a model or two, thereby letting the customer make an informed choice. Salespeople emphasize that at this stage, as in Stage IV, model recommendation is not based on the price range customers consider but on preferences for certain generic or specific features.

In response to the customer's request, the salesperson presents comparative information in relation to models of other manufacturers that are believed to compete with the model in question. It is more difficult to present comparative information in Stage II because of wide variation in price and features availability among models of a given carline.

VII. Minor Optional Features: Provide Information, Identify Preferences and Needs, Recommend, and Let Customers Select

Unlike the choice of a model, which is typically driven by preferences, the choice of options may be driven by both preferences and needs. Salespeople develop certain heuristics for option recommendations and information. While some options are always recommended (e.g., tinted glass, a light in the trunk), other options are recommended only to certain persons (e.g., tilt wheel for heavy, tall, or short people; security features to women). Salespeople also emphasize the importance of explaining/showing the benefits of optional features to the customer.

The recommendation of an optional engine is typically determined by price and other factors like the customer's preference for power (e.g., 0-60 mph acceleration), the perceived importance of fuel efficiency, and the expected annual mileage. The latter two factors are interrelated because for many customers the higher the annual mileage, the more important fuel efficiency becomes.


In this section we briefly compare our model of the customer-salesperson interaction to the findings of two other studies. One examined the role of salespeople in the purchase of major appliances, including automobiles. The other investigated salesperson effectiveness.

The Role of the Salesperson

Wilkie and Dickson (1985) and Dickson (1986) report the results of an extensive study of customer shopping for major appliances and for automobiles. They conclude that salespeople often become the dominant source of influence in a customer decision. As seen in Figure 2, once the customer enters the store, the role of the salesperson is dominant in two significant functions: to assist customers in "Purchase Respecification," and in "Identification of Best Alternative."

Although Wilkie and Dickson's model is general and goes beyond the salesperson-customer interaction, while our model encompasses only the dealership dialogue, the two models assign the -salesperson similar roles. The role of the salesperson in "purchase respecification" is elaborated in our model by a sequence of tasks: identifying preferences, providing information, identifying needs, and reconciling preferences and needs.

With respect to the "Identification of Best Alternative," the first four stages in our model prepare the ground for the major customer decision of carline selection in Stage V. The salesperson encourages the customer to go back and revise the information search (Stage II), consider salesperson's recommendation (Stage IV), and eventually in light of new information, change major preferences (Stage I), to converge towards the "Best Alternative." In addition, because of our product hierarchy, the identification of best alternative is spread out also over stages VI and VII.

The implication of Wilkie and Dickson's cost/benefit analysis (by the customer) for "further search" (see Figure 2) is that the salesperson should have the capability of making instant comparisons with products of other makes (end of Stage VI in our model). In this way the customer benefit of in-store search might be larger and the benefit of external search (e.g., identify other makes or stores) would become smaller, in comparison to the implied cost.

Sales Effectiveness

A study by Weitz et al. (1986) centers on the ability of salespeople to adapt effectively to the demands of the sales situation. Adaptivity is considered crucial because it indicates the degree to which salespeople are able to take advantage of the unique communication elements associated with personal selling (i.e., messages should be adapted to the specific customers). Using a categorical model of memory developed by Rosch and her colleagues (e.g., Rosch and Mervis, 1975), Weitz et al. propose that the effectiveness of adaptive selling increases with:

(1) number of sales- situation categories available in the salesperson's long-term memory;

(2) degree to which salespeople have a hierarchically organized knowledge structure;

(3) degree to which salespeople classify sales situations in terms of underlying characteristics, such as the effect of different sales approaches; and

(4) level of procedural knowledge (in contrast to factual knowledge).

Our salesperson-customer interaction model is consistent, in general terms, with the above four propositions on adaptive selling. However, the principle of adaptivity does not match in any detailed way the sequential dialogue that we propose. Specific hypotheses would have to be derived for the automobile dealership setting. Alternatively, the various propositions of Weitz et al. about the salespeople's knowledge structures could be tested. It would be interesting to test both ESPRI and individual salespeople for their capacity to adapt to different customers. We discuss the issue of model testing in the following section.


Recent knowledge engineering literature points to the importance of modeling a process (be it medical, engineering, social, or psychological) in the development of an expert system (e.g., Clancey and Rothenberg, 1987, p. 159). Until recently, the accent in expert system development was on modeling one expert rather than modeling the process of interest. In the present study we modeled the customer- salesperson dialogue, by integrating the expertise of a group of salespeople. We separated a core customer-salesperson interaction from the idiosyncratic approaches that individual salespeople have developed over time. The resulting model is currently being used as the basis of ESPRI, an expert system for aiding car buyers (and, possibly, salespeople).

Once ESPRI is sufficiently completed, the challenge is testing, diagnosis and revision. That is, an ESPRI that genuinely reflects the potential of knowledge-based systems technology will emerge only through an iterative process. The system should be tested on (preferably) real car buyers in dealer showrooms. The point where customers cease using the system, why they quit, what information they requested that the system was unable to provide, and so on, can be identified. This feedback should enable a diagnosis of system weaknesses and point the way to revision.

The testing of the model that is the focus of this paper is implicit in the testing of ESPRI. That is, ESPRI may fail because the content or order of our seven stages is wrong -- or wrong for some customers -- or for non-model reasons such as lack of user friendliness, slowness of responding, etc. The main point is that the model is part of a larger system and will be tested within tests of that system. Although such testing may require a significant research effort, it is also complete and rigorous. From it a validated model of the customer - salesperson interaction and an effective ESPRI should emerge.




Clancey, William J. and Jeff Rothenberg (1987), "Evaluating Expert Systems Tools," Sixth National Conference on Artificial Intelligence, Conference Tutorial Program, (July), Seattle, Washington.

Dickson, Peter R. (1986), 'The Purchase Participation Roles of the Salesperson in the Buying of Automobiles and Major Durables: A Survey Research Brief Report," WPS 86-54 (April), College of Administrative Science, Ohio State University, Columbus, Ohio.

Harmon, Paul and David King (1985), Expert Systems, John Wiley and Sons, New York.

Rosch, Eleanor and Carolyn B. Mervis (1975), ''Family Resemblance: Studies in the Internal Structure of Categories," Cognitive Psychology, 7 (October), 573-605.

Weitz, Barton A., Harish Sujan, and Mita Sujan (1986), "Knowledge, Motivation, and Adaptive Behavior: A Framework for Improving Selling Effectiveness," Journal of Marketing, 50 (October) 174-191.

Wilkie, William L. and Peter R. Dickson (1985), "Shopping for Appliances: Customers' Strategies and Patterns of Information Search," Working Paper 85-108 (November), Marketing Science Institute, Cambridge, Mass.