Short Programs

Data-driven Marketing: Pricing, Bundling, and Customer Targeting

Date: TBD, 2011 | Tuition: $1,500 (tentative) | Continuing Education Units (CEUs): 0.9 (tentative)

Overview  |  Background  |  Learning Objectives  |  Pre-Requisites  |  Topics Covered  |  Schedule  | 
Staff  |  Location  |  Updates

Overview

Participants will spend 2 intensive days learning about how a firm can create innovative promotions that will achieve strategic objectives utilizing historical data sets. A firm's objectives range from maximizing profit or revenue to maximizing market share in the presence of competition. Currently, firms attempt to solve this problem by employing disparate techniques from data mining and predictive modeling. However, a usable consistent methodology for promotion design is still lacking. This course will present a comprehensive approach for optimal promotion design, which integrates state-of-the- art data analytics, marketing science, and optimization.

Marketing, merchandising, product managers, statisticians and data analysts will find this course invaluable.

BACKGROUND

One of the most effective ways a company can increase profits is by selling the right products/bundles at the right prices to the right people. A promotion is effectively a limited-time price reduction, and therefore, promotions have significant costs that can make a big dent in the bottom-line. Companies face significant challenges in designing promotions, and money is inevitably left on the table. Some of the challenges include:

• Rejected Offers: a majority of offers are still not relevant to most prospects (95-99% of offers are rejected),
• Eroding Profits: promotions reduce margins without driving behavioral change (10% discount can reduce profit margins by > 50%),
• Limited Learning: promotion analysis teaches a manager whether or not to repeat a past promotion, but does not help in developing better promotions.

The fundamental idea behind successful promotion/offer strategies is to best understand the customers' utility and behavior. The customers' utility captures their specific interests in the product line, their price sensitivity with respect to specific product/service, and their budgetary constraints. The behavior captures their frequency of purchase and amount spent per transaction. With this information, the firm can design their strategy to maximize its objective function.

Historical data can be used to model the customers' utility and behavior. Typically, it is not possible to generate consistent models that describe each customer's utility and behavior; however, such information may still exist at an aggregate level. Segmenting the customers to derive such models is the first and most basic challenge in this process. The optimal segmentation provides groupings of customers such that each group has similar utility and behavior and is distinguishable from other groups. Essentially, such a segmentation generates homogeneous sub-markets from a heterogeneous market.

Once the customers' utility and behavior is understood, a model of how the customers' utility and behavior may change with respect to a new offer /promotion needs to be developed. Typically, these are called "impact models". For each specific new offer template, a model can be hypothesized that describes this change, and based on this model, the different parameters of such templates can be optimized. For example, given a business objective, marketing constraints, and promotion templates such as "Buy product X and get product Y at R% off" with unknown parameters X,Y,R,Z and Q, the problem of promotion design can be stated as: find the optimal set of parameters for a finite set of promotions that reach out to a significant number of customers. Given an impact model, the search for these parameters can be prohibitively complex. Tractability of any marketing process is a critical piece of any successful methodology.

This course will take you through each step of the promotion design process and will discuss how ad-hoc approaches used for each step create unacceptable errors, which cost businesses substantial money. A comprehensive approach for each step will then be covered.

LEARNING OBJECTIVES

1. Evaluate marketing’s role in surviving in a competitive environment: how to frame strategic vs. data driven decisions, games and price wars, and competitive positioning
2. Analyze global modeling across channels: how to find and exploit distinguishable submarkets within a global market
3. Assess optimal promotion design which integrates advanced data analytics, marketing science, and large scale optimization
4. Examine data consolidation: how to identify and extract historical data sets that can be reliably used to optimize promotions with respect to a given business objective
5. Summarize how a firm can create innovative promotions which induce behavioral change in customer purchase patterns
6. Compare and contrast current modeling approaches: clustering and modeling techniques such as k-means clustering, regression models, artificial neural networks, decision trees, machine learning, and other techniques.
7. Measure the effects of advertising vehicles by targeted segments: identify impact models and demand lift associated with various promotional media and offers
8. Design continuous learning engines: utilize iterative feedback mechanisms which continuously update marketing models based on new strategies as well as shifts in market environment and response.
9. Compare the issues and challenges of real life example firms.

PRE-REQUISITES

Basic marketing concepts and data modeling are helpful but not necessary.

TOPICS COVERED

1. Data Consolidation: How does one extract data relevant to optimizing promotions with respect to a given business objective?
2. Current Modeling Approaches: What models do people build today? Is a query a model? Why should marketing people care about models? Do all models provide the same utility? An overview of clustering and modeling techniques such as: k-means clustering, regression models, artificial neural networks,, decision trees, machine learning, and other techniques will be provided.
3. Global Modeling: How does one find distinguishable submarkets within a global market? Modeling customers' utility.
4. Marketing Science: How do we innovate promotions? What impact models exist for various promotions? Does tracking help in successive iterations.
5. Surviving in a Competitive Environment: Modeling competition. Games and price wars. Strategic decisions vs. data-driven decisions.
6. Real-life Examples: Four real life examples will be presented:

• book retailer: issues and challenges
• computer retailer: issues and challenges
• clothing retailer: issues and challenges
• furniture retailer: issues and challenge

Course schedule and registration times

Class runs 9:30 am - 5:00 pm on Monday and 9:30 am - 3:00 pm on Tuesday.

Registration is on Monday morning from 8:45 - 9:15 am.

STAFF

Munther A. Dahleh
Munther is a recognized leader in systems modeling, learning, and robust control. He has been a Professor in the EECS Department at MIT, leading the systems and control group, since 1987. He is also the author of two textbooks and numerous papers in the areas of feedback control and learning. He has also been a visiting Professor at the California Institute of Technology, has held consulting positions with major global corporations, and was a co-founder of Crescent Technologies, a company that focused on supply-chain integration with production. Munther is also a co-founder of Infolenz Corporation. InfoLenz was born as a group of MIT scientists identified significant limitations in conventional analytics techniques, which apply ad hoc statistics and optimization rather than focusing on the true underlying economics of markets. Munther led the team, which developed a patent-pending methodology that helps companies leverage their market economics to drive sales.

He completed his Ph.D in Electrical Engineering from Rice University in 1987 and received the Ralph Budd award in 1987 for the best thesis at Rice University and the George Axelby outstanding paper award (paper co-authored with J.B. Pearson in 1987). He also was a recipient of the NSF Presidential Young Investigator Award in 1991, the Finmeccanica career development chair in 1992, the Eckman award for the best young control engineer in 1993, and the Graduate Students Council teaching award in 1995. He is a fellow of IEEE.

Sridevi Sarma
Sridevi is also a cofounder of Infolenz and is currently a post doctoral fellow at Harvard focusing her research on applications of large-scale modeling to the understanding of the nervous system. She completed her Ph.D. in 2006 in the department of Electrical Engineering and Computer Science at MIT. She has taught numerous undergraduate courses at MIT in probability theory, signal processing and systems and signals. She has also taught graduate level courses in dynamic systems. Her research examines the modeling of complex, large-scale distributed systems, most notably the brain, using control systems theory. Her work in this area has far ranging impact on the modeling of complex, distributed systems such as e-commerce hubs on the Internet. She received her MS in the Laboratory of Computer Science at MIT, during which she gained expertise in speech recognition technology. She applied this knowledge to a project involving phonetic classification at Digital Equipment Corporation. She is a former National Science Foundation fellow and a recipient of the GE Faculty for the Future scholarship.

Location

This course takes place on the MIT campus in Cambridge, Massachusetts. We can also offer this course for groups of employees at your location. Please complete the Custom Programs request form for further details.

Updates

This class is tentatively planned for 2011, depending on the level of interest. Email the Short Programs office to express your interest in taking this course. Please include your industry and learning goals.