Tutorial: Optimizing Customer Loyalty

Stefan Conrady; Lionel Jouffe

Tutorial: Optimizing Customer Loyalty

An Integrated Market Research Workflow Using Bayesian Networks and BayesiaLab

Executive Summary

Identifying Priorities for Maximizing Repurchase Intent

This tutorial illustrates an innovative market research workflow for deriving marketing and product planning priorities from auto buyer surveys. In this study, we utilize the Strategic Vision New Vehicle Experience Survey, which includes, among many other items, customers’ satisfaction ratings with regard to over 100 individual product attributes.

Challenge: Indistinguishable Drivers of Loyalty

With traditional statistical methods, it has been difficult to rank the importance of individual product attribute ratings with regard to an overall measure, such as repurchase loyalty.

The key challenge is that customers’ ratings of individual product attributes are highly correlated. When plotted, we see 100 lines that are nearly indistinguishable in terms of their slope. Given this collinearity of all variables, traditional statistical methods fail to distinguish the importance of individual ratings. We could only naively conclude that an improvement in any rating would generally be associated with higher loyalty. No clear priorities could be established on such a basis.

Solution: Bayesian Networks as Modeling Framework

To overcome this problem, we employ an alternative framework: we use Bayesian networks as the mathematical formalism, plus the machine-learning and optimization algorithms of the BayesiaLab software package. This approach embraces collinearity as a feature in the model, instead of suppressing it as a nuisance.

Implementation

First, using BayesiaLab, we machine-learn a Bayesian network that models customers’ brand loyalty as a function of their ratings of their current vehicle. This identifies key factors as loyalty drivers in the overall market, at the segment level, and finally at the model level. With these factors identified, we perform optimization for each vehicle within its competitive context. As a result, we obtain a list of specific priorities for each vehicle, along with the simulated gain in loyalty.

Other Benefits

No Black Box

Many modeling techniques offered in the field of marketing science are opaque to the end user of the research. The nature of many models makes them inherently black-box, and thus requires a leap of faith by the decision-maker.

Not so in our research framework with Bayesian networks. Regardless of one’s quantitative skills, any subject matter expert can, by simply using common sense, interpret the Bayesian network models generated with our workflow. Any stakeholder can immediately scrutinize such a model, thus enabling him to verify its structure, or, by using his domain knowledge, to invalidate it. Their inherent falsifiability makes Bayesian networks ideal scientific tools.

Real-Time Recommendations

In most organizations, waiting for research results and their interpretation is a matter of months. The time span between a consumer sentiment expressed in a survey and a company’s response can sometimes even exceed the lifecycle length of a product.

Our workflow creates a single, direct, and transparent link from data to recommendation. This directness provides unprecedented analysis speed. We reduce the lag between receipt of data and delivery of recommendations from months to days. As a result, near real-time policy recommendations are feasible for the first time.

Introduction

Background

Market Maturity and Homogeneity

The auto industry is an example of a mature market. It is fair to say that all automakers offer high-quality products these days in North America. The proverbial “lemons” are few and far between. Fierce competition has led to product offerings that are remarkably similar for their respective vehicle category, both in their specifications and their functional performance. With similar cost and budget constraints, and an overlapping supplier base for all manufacturers, the auto business is mostly about eking out minute advantages, as opposed to creating fundamental breakthroughs.

No doubt, the brand plays a major role in buying decisions. Hence, marketing, branding, and promotion efforts of automakers typically absorb a similar amount of resources as the actual R&D expenses for vehicle development. For the purpose of this paper, however, we will not venture into the challenging domain of return on marketing investment. This is a topic for another methodology tutorial in the future.

Given this overall quality and performance homogeneity, consumer perceptions, as we will see in this study, are also remarkably homogeneous across similar kinds of vehicles. For market researchers, it is thus very difficult to “tease out” material differences in customers’ perception of product attributes of competitive vehicles. It is even more challenging to establish which of these similarly-perceived vehicle characteristics do really matter when it comes to buying an automobile.

Loyalty

“It is cheaper to keep a customer than to find a new one” is an often-quoted marketing adage. Loyalty is a very relevant quantity, much more tangible than mere satisfaction. Given the maturity of the auto market and rather lengthy ownership cycles, repurchase loyalty is of special significance. Thus, in this study we go beyond satisfaction and instead link product ratings to stated repurchase intent.

We will focus exclusively on how customers’ product ratings affect loyalty, i.e., what really matters for brand loyalty. We will present a methodology that identifies the relevance of minute differences in consumer perception to prioritize among a broad range of opportunities to improve product ratings.

Workflow Overview

Latent Factor Induction

In this paper, we employ BayesiaLab’s machine learning algorithms to generate Bayesian networks that will allow us to identify major concepts, i.e., latent factors, from the observed satisfaction ratings, i.e., manifest variables. Inducing factors creates a level of abstraction that will allow us to see a “bigger picture,” that is more stable than if it is only based on manifest variables. Once factors are identified, we will examine how they “drive” brand loyalty. Ultimately, we want to establish the effect of these factors with regard to the outcome variable, i.e., loyalty.

Multi-Level Analysis

We will examine loyalty drivers at multiple levels of the market. We will identify areas of opportunity for improving loyalty at both the segment level and the vehicle model level.

Segment refers to a vehicle category, such as Subcompacts or Large Sedans. There are numerous segmentation schemes in the auto industry, each with its own terminology. However, automakers generally agree on the definition of the Full-Size Pickup segment, which is the focus of this study.
Vehicle model refers to a make (brand) and model/line, e.g., Ford Explorer or Nissan Altima. In this case study, we do not drill down to the trim level, e.g., Ford Explorer Limited or Nissan Altima 2.5 S.

More specifically, we will proceed from the overall market to the Full-Size Pickup segment, and then to the vehicle models within it. We chose this particular segment primarily for expository simplicity. It is a very well-defined segment in terms of vehicle characteristics while consisting of only a few major contenders. Plus, it is one of the most important segments in the U.S. auto industry, both in terms of volume and profitability.

Optimization

Once loyalty drivers are modeled, we will identify priorities for improvements by vehicle model. For each model, in its specific competitive context, our approach will generate recommendations with the objective of improving brand loyalty.

As we examine the impact of satisfaction ratings on loyalty, we need to remember, though, that satisfaction ratings are inherently subjective. The recommendations we will present do not necessarily specify the means by which ratings should be improved.

Acknowledgements

We would like to express our gratitude to Alexander Edwards, President of Strategic Vision, Inc. [3], for generously providing data from their 2009 New Vehicle Experience Survey for our case study.

Notation

To clearly distinguish between natural language, software-specific functions, and example-specific variable names, the following notation is used:

Exact BayesiaLab commands, menu paths, and GUI labels are shown in backticks.
Menu paths use the > separator.
Example-specific names retain their exact capitalization when referenced directly.

[3] Strategic Vision is a research-based consultancy with more than 35 years of experience in understanding consumers’ and constituents’ decision-making systems for a variety of Fortune 100 clients, 10 Downing Street, Coca-Cola, American Airlines, Procter & Gamble, the White House, and most automotive manufacturers and many advertising agencies. The company specializes in identifying consumers’ complete motivational hierarchies, including product attributes, personal benefits, values/emotions, and images that drive perceptions and behaviors. Strategic Vision has at its core a large-scale syndicated automotive experience and “Pulse of the Customer” (POC) study that collects more than 350,000 responses annually, using over 1,500 comprehensive data points. Since its foundation in 1972 and incorporation in 1989, Strategic Vision, led by company founders Darrel Edwards, Ph.D., J. Susan Johnson, Sharon Shedroff, and Alexander Edwards, has used in-depth Discovery Interviews and Value Centered Survey instruments that provide comprehensive, integrated, and actionable outcomes, linking behavior to attributes, consequences, values, emotions, and images (www.strategicvision.com ).

Sections

The tutorial is presented in seven parts:

Summary

Bayesian networks and BayesiaLab make it possible to identify meaningful drivers from previously indistinguishable product ratings in survey data. On that basis, BayesiaLab can perform optimization and immediately establish priorities. With this approach, market researchers can quickly and transparently generate clear recommendations for decision-makers.

Seminar: Key Drivers Analysis and Optimization Data Preparation