Predicting Customer Choices

Vanishing mass markets, the proliferation of products and services and new technologies are requiring many companies to redefine their beloved core business doctrine: “Give customers what they want.” At the same time, in the best-selling book The Paradox of Choice, Schwartz (2004) suggests that every consumer decision, from buying a bottle of shampoo or ordering a cup of coffee to choosing a healthcare provider or setting up a retirement plan, is becoming increasingly complex thanks to the abundance of choices available. Much the same is true for customers in many business-to-business markets.

This dramatic explosion in options has ironically become a challenge for customers and businesses alike; Schwartz goes so far as to argue that fewer alternatives are better than many for the well-being of society. However, the underlying problem in predicting customer choices resides largely in the fact that many people make purchasing decisions on the basis of (potentially) many different criteria simultaneously (McFadden 1986), including brand, quality, performance, price, service, features, channel and so on.

Given resource constraints, it is virtually impossible for any firm to excel in all product aspects at once — that is, to provide the highest quality, fastest delivery and greatest variety at the lowest price. Firms must make trade-offs on the basis of what they do best, what their competitors are offering, and the criteria they think matter most to their customers (Verma, Plaschka and Louviere 2002). However, managers often struggle to determine the “best” configuration of product-service offerings that will appeal to their chosen target markets. To create, capture and maintain demand for their offerings, businesses have to balance three key challenges (Verma and Plaschka 2003).

Ambiguity — What do our customers really want? Companies lacking a clear understanding of customer choices often take a scattershot approach, hoping that at least one of their offerings will succeed. Unfortunately, thistype of approach is neither efficient nor profitable for most firms. Markets are frequently flooded with products and services that offer relatively little in the way of added value to customers and that weaken the seller’s bottom line.

Risk — Will our envisioned offerings be successful? Managers face complex choices when deciding which of their product-service bundles to offer. Potential product-service drivers (such as price or specific product-service features) can have several variants, and managers often use experience, benchmarking analysis or simply gut feelings to decide what will be attractive to customers. Such “informed guessing” may spur new and innovative ideas, but it also may be unproductive and unreliable.

Conformance — Can we deliver what we promised? Although it is important for companies to understand market value drivers, they must also support customer preferences and align them with effective supply-chain management practices. Even if companies succeed in identifying and delivering attractive product-service packages, their efforts may prove futile unless they can efficiently deliver on their promises under resource constraints.

The Science of Choice Modeling

During the last few years, new research has expanded the toolbox that is available to businesses seeking to understand the drivers of customer choices. The latest tools and methodologies allow managers to predict with remarkable precision the market performance of new or existing products, services or experience-based offerings. Recent works on the “art and science” of choice modeling approaches have greatly improved management’s ability to predict customer choices — even under seemingly complex and erratic market conditions.

The CM framework pioneered by Daniel McFadden, corecipient of the 2000 Nobel Prize in economics, focuses on both the economic reasons for individual choices and the ways researchers can measure and predict these choices (Manski 2001; McFadden 1986). McFadden’s work, and corresponding developments in experimental choice analysis by Jordan Louviere and his co-researchers, have led to diverse applications, including design and development of new products and services, transportation planning in urban environments and evaluation of alternative pricing strategies (Louviere and Woodworth 1983; Adamowicz et al. 1998; Ben-Akiva and Lerman 1985; Louviere, Hensher and Swait 2000). For example, in a recent article, Hall et al. (2004) describe the use of CM for a variety of managerial applications in the healthcare industry; Verma, Iqbal and Plaschka (2004) do much the same for the online financial services industry.

The Art of Choice Modeling

Experimental choice modeling requires customers (respondents) to make choices in simulated situations derived from realistic variations of expected market offerings. The process typically involves three broad steps.

The first step is built around understanding the range of possible choices (or metachoices) of the target area. Using qualitative market assessment, customer interviews, case studies, industry data, focus groups and other information sources, companies compile a list of market drivers that they believe influence customers’ buying decisions. For example, a residential housing developer might identify choice drivers, such as an apartment’s size (onebedroom, two-bedroom, three-bedroom); kitchen layout (L-shaped, U-shaped, galley style, open shape with island); appliances (standard, designer, professional); amenities (exercise facility, roof terrace, theater/ entertainment suite, concierge/doorman); parking (valet or self-parking, car washing/ cleaning services); and price factors (base price, built-out cost, floor location, views and amenity pricing).

The second step is to design experiments that ask respondents to select one out of two or more choice options in a series of “choice sets.” Verma, Iqbal and Plaschka (2004) presented respondents with descriptions of two online financial services packages in a series of 16 different choice sets. Within each set, respondents were asked to select either one of the two packages or neither. Choice tasks can be constructed in many variations, as described in Louviere, Hensher and Swait (2000) and Train (2003).

The third phase involves using econometric models based on responses from a representative sample of customers (or potential customers) to identify empirical key patterns in the survey responses, relative weighting for each market driver and for interactions among drivers. Managers can select the optimal combination of operations and market drivers in an effort to develop the most profitable and sustainable value proposition that leverages available resources as much as possible.

New Advances

Like any science, choice modeling continues to evolve as researchers from various academic and professional disciplines pursue projects with diverse focuses and emphases. At the same time, the “art” of CM is also evolving as information technology makes it possible to develop more elaborate and realistic choice experiments. Some of the recent advances are as follows:

Realistic Experiments Only a few years ago, a typical implementation of choice modeling involved having respondents react to lengthy paper-and-pencil surveys based on a series of preconfigured choice scenarios. Choice sets were presented as static tables with little room for customization. Recent advances in experimental designs and IT, including broadband Internet access, digital imaging and video, and faster computing speeds, have allowed researchers to develop more realistic choice experiments that can be adjusted to an individual respondent’s decision scenario. In recent work across a wide range of industries (retail, hospitality, financial services, industrial automation, medical systems, telecommunications), Web-based technologies (hyperlinked pictures, brand logos, audio and video files) have been used to illustrate the choice scenarios (Verma and Plaschka 2003, 2005).

Advanced Procedures While the role of information technology in designing realistic experiments is impressive, even more impressive are the contributions of statisticians and management science researchers who have been developing advanced procedures for estimating and fine-tuning econometric models on the basis of choice modeling. Advances in Bayesian statistics allow researchers to estimate choice models for each individual respondent and/or finetune market-segment membership. Several such statistical advances are described in a recent book by Train (2003). Innovative optimization procedures involving chaos theory, neural networks, simulated annealing, genetic algorithms and simulation modeling are being used in various applications to identify optimal product-service design configurations and to link choice modeling outcomes with other managerial decision problems (for example, Bonabeau 2002). Other advances in choice experiment design include developing hierarchical choice experiments and partial profile designs. While such procedures increase the complexity of designing choice studies, data analysis and econometric model estimation, they also allow researchers to reduce the choice-task complexity for respondents by showing only a few market drivers at a time within each choice set.

Data Fusion During the last few years, firms have invested heavily in customer-relationship management systems and IT in general. Such implementation generates a huge amount of customer transaction data (such as airline or hotel check-in records, reservation patterns, usage reports on various facilities, credit-card usage patterns, frequent user/loyalty card records, wireless voice and data records), which can be studied to analyze customer preferences over a long time. Effective use of CRM data can allow organizations to customize product-service offerings based on usage patterns of individual customers, thereby increasing satisfaction, retention and loyalty (Loveman 2003).

Although the use of CRM and data mining techniques can be extremely helpful in isolating trends based on past choices, such approaches can have only limited applicability when making predictions about really new product-service features. Hence, choice modeling results, together with econometric models drawn from existing CRM databases, can at best only estimate the impact of innovations within a particular business context (Hensher, Louviere and Swait 1999; Swait and Andrews 2003). As a result, one needs to be extremely cautious about using such data-merging techniques, in order to isolate any statistical differences that may occur from using multiple methods; otherwise the resulting models may be confounded by random errors (Louviere 2001).

Switching Inertia Many customers are reluctant to give up their current productservice provider for one or more reasons: a habit or preference for the status quo (“don’t like to switch”); satisfaction with current product and service offerings; lack of real or perceived alternatives; lack of credible alternatives; and customer loyalty. However, the reluctance to switch (inertia) does not account for such factors as lack of awareness, contractual limitations, ignorance of alternatives and paradigm shifts (for example, product obsolescence). Robust and reliable estimates of switching inertia can be derived by designing choice experiments in which respondents choose between “current” and “new” product-service providers (Li et al. 2005). Thus, companies can see the value of “cash-like incentives” to acquire customers and avoid the “customer reversal process,” in which customers return to their former base because the company’s offering has no value beyond the initial inertia-breaking incentive (such as a switch between wireless carriers for free phones; a switch between hotel or airline operators for free loyalty points).

Customer Satisfaction — Experience Assessment In practice, most firms use a series of rating questions (for example, a 10point scale ranging from “least satisfied” to “most satisfied”) to measure customer satisfaction on predefined criteria. However, respondents are notorious for rating items too rapidly, using simplification heuristics to speed through the task (Cohen and Orme 2004). Many respondents use only a limited range of the scale points, resulting in many ties among items. Some respondents use just the top few boxes of a rating scale, some refuse to register a top score for any item and others spread their ratings across the entire range. To overcome these problems, Finn and Louviere (1992) have developed an approach known as “best-worst choice analysis,” which allows researchers to develop more robust segment-level predictive models free from scale biases. In this approach, the respondent is asked to identify best and worst features presented on a latent dimension (for example, attractiveness or satisfaction). Compared with traditional rating scales, the derived best-worst scores are based on the relative comparisons among the items in the study, with a significantly better “signal-to-noise” ratio. The ability to distinguish across items permits researchers to uncover differences among the various respondents effectively. This development is a significant improvement over common practice that typically uses rating scales as the basis variables for customer satisfaction measurement or market segmentation.

Conclusion

Choice modeling can yield valuable insights for market-science– and decisionscience–driven strategy development by revealing customer-needs–based segmentation maps, by measuring the market share impact of alternative or new product-service feature combinations and by assessing overall brand equity in the context of identifiable switching barriers. Moreover, choice modeling can reveal salient differences between managers’ beliefs about customers’ needs and wants and customers’ actual choices. For managers seeking reliable feedback on how customers view their offerings, CM provides a rigorous method of turning customer-driven feedback into profitable and sustainable strategies for retaining or capturing market share.

At the same time, choice modeling, like other modeling processes, is subject to the principle of “garbage in, garbage out.” It generates useful information only if the assumptions behind the selection of market drivers, the experimental design and the data collection methods are sound. At its best, CM offers powerful tools for developing more attractive offerings, which form the basis of creating winning business strategies and market-share growth without sacrificing margins.