• The best seller "Supercrunchers" by Ian Ayers (Random House, 2007) has introduced analytics to a wide audience.
  
  
• "60 Minutes" on CBS television recently had a piece on Bill James and the analytics he has brought to the Boston Red Sox (http://www.cbsnews.com/stories/2008/03/27/60minutes/main3974752.shtml.)
  
  
• An article in Canada's major newspaper in January listed eight big trends to think about in preparing for 2008 and beyond. The first of these was: "Revenge of the Math Geeks - Analytics is hot. Those kids back in high school who were really good at math own the future. Whether it's balancing loads in energy systems or decoding our DNA, the 21st century is all about math. Its where the next billion-dollar industries will be born." Globe and Mail, Jan. 7.
  
  
• Business Week published an article on a similar theme: "A generation ago, quants turned finance upside down. Now they're mapping out ad campaigns and building new businesses from mountains of personal data." ("Math Will Rock Your World" by Stephen Baker, BusinessWeek. Jan. 23, 2006)
  
  
• Following the successful use of analytics in baseball ("Moneyball: The Art of Winning an Unfair Game" by Michael M. Lewis; W.W. Norton & Co., 2004) and "Royals' Bannister unafraid to do the math" by Jeff Passan, Yahoo! Sports, March 8) we Canadians are pleased to see that analytics has even found its way into hockey ("Number cruncher courts Canucks" by Iain Macintyre, Vancouver Sun, May 6.
  
  
• INFORMS has started publishing the e-zine, Analytics (www.analytics-digital.com/analytics/2008spring/?pg=4)

My research lead me to the fact that "analytics" has been around for some time. Wikipedia suggests that "the simplest definition of Analytics is 'the science of analysis,' " which I don't think is too helpful. I have found out that "analysis" means different things to different people; in particular "analysis" can be entirely qualitative (I have heard the use of the SWOT [strengths, weaknesses, opportunities and threats] framework described as rigorous analysis!) Wikipedia goes further, adding, "a simple and practical definition, however, would be how an entity (i.e., business) arrives at an optimal or realistic decision based on existing data. Business managers may choose to make decisions based on past experiences or rules of thumb ... but unless there is data involved in the process, it would not be considered analytics." This later statement suggests that decision-makers practice analytics when they consult data before making a decision — pretty close to a definition of O.R.

The relationship between O.R. and analytics, however, is not at all clear. In his Harvard Business Review article (January 2006), "Competing on Analytics," Thomas Davenport describes one successful O.R. application after another but seems to go out of his way to avoid using the term "operations research." He suggests that "Proctor & Gamble, for example, recently created a kind of beranalytics group consisting of more than 100 analysts from such functions as operations, supply chain, sales, consumer research and marketing," although most of us would see this as an O.R. group in the tradition of Harold Larnder's operational research group at fighter command in 1938.

Davenport's view of analytics appears to be consistent with a view that O.R. is a subset of analytics (perhaps "turbo-analytics"), but there is a body of skills and knowledge that is analytics but not O.R. This raises interesting questions: What is this body of knowledge and who teaches it?

Wikipedia's definition of business analytics is "how organizations gather and interpret data in order to make better business decisions and to optimize business processes," suggesting that "gathering and interpreting data" might be analytics but not O.R. The "60 Minutes" segment on Bill James and the Red Sox spends a lot of time talking about new metrics to measure performance and how the use of these metrics can help the manger make better heuristic decisions. There is a recurring theme that a body of data analysis skills exist that are uniquely analytics: "Common applications of analytics include the study of business data using statistical analysis in order to discover and understand historical patterns with an eye to predicting and improving business performance in the future," and "analytics closely resembles statistical analysis and data mining, but tends to be based on modeling involving extensive computation." (Wikipedia).

Davenport and Jeanne Harris in their book "Competing on Analytics: The New Science of Winning" (Harvard Business School Press, March 2007) define analytics as "the extensive use of data, statistical and quantitative analysis, explanatory and predictive modeling, and fact-based management to drive decisions and actions. Analytics may be used as input for human decisions; however, in business there are also examples of fully automated decisions that require minimal human intervention."

Three Skill Sets


Recurring points in these definitions appear to segment analytics into three skill sets. One is the grunt work that O.R. people have to do before they can apply O.R. methods and models. This includes data gathering, data cleaning and organizing, distribution fitting, descriptive statistics, "trend spotting" and developing performance metrics. I will call this body of knowledge and skills "pre-O.R." The second piece of analytics is the O.R. skill-set itself, which would include building some form of mathematical or spreadsheet model to add value to the data in helping understand and resolve a decision. The third segment of analytics consists of a set of more advanced analytical skills and methodologies that would not normally be taught in an O.R. course. These include data mining, predictive modeling and advanced statistical techniques such as logit, cluster analysis and some tools from artificial intelligence, as well as methodologies such as Six Sigma and lean manufacturing. I will call this set of skills "non-O.R. analytics."

When we look at how we teach analytics, I think we do a good job teaching the O.R. segment. Some schools have courses in the non-O.R. analytics areas, presumably the result of both student demand and faculty teaching interests. Other methodologies (for example: Six Sigma) have extensive educational structures in place and there is, at best, only a peripheral role for O.R. My concern in this article, however, is how and where our students learn the pre-OR analytics skill-set. Where do they learn how to do the grunt work that precedes every real O.R. modeling exercise? There is not a lot of point teaching how to formulate and solve a mathematical program if the student does not know how to collect the relevant data, clean it and organize the data into a form that will allow the optimization to work. Similarly, queueing theory can provide a host of valuable insights, but its use requires gathering arrival and service times, fitting their distributions and (usually) calculating some summary statistics.

My experience, based largely on a great many EMBA projects, is that there is often an 80/20 rule of analytics: 80 percent of the effort often goes into the pre-analytics grunt work, but once this is done the O.R. modeling is often straightforward. Further, collecting data, cleaning it, organizing it and running some descriptive statistics is often enough for an intelligent decision-maker to sufficiently improve their decision-making that they don't feel that they need to bother with the model.

If pre-O.R. analytics is so important, we should teach our students how to do this. I have looked at many O.R. course outlines and I am concerned that we often outsource teaching the grunt work to our colleagues in statistics departments. In many schools, O.R. class time appears to be too important to devote to material that is not strictly O.R. Many business programs have a required statistics course and a required O.R. course (although these may not be the labels used), and so we may expect (or hope) that our colleagues teaching statistics cover pre-O.R. analytics, but this may be a stretch. Among the broad range of statistics courses, some are focused on decision-making with data, but how many teach students how to design a survey instrument, how to find a distribution that approximates a set of real data, or how to dredge really useful data from the mess that is often the corporate data base?

How to Teach Pre-O.R.


Given we cannot rely on our colleagues to teach pre-O.R. analytics for us, it seems obvious that if we are trying to produce graduates who can actually do O.R., we need to teach these materials ourselves. How do we do this?

The obvious approach is to have students do real-world O.R. The real-world project must be the most effective way to learn many of the different skills in the pre-O.R. skill set. Many schools execute student projects really well; for example the COE program at University of British Columbia and the O.R. program at Lancaster University Management School. These examples, however, illustrate the kind of commitment necessary to use student projects effectively. There needs to be full-time staff whose job is to find the projects, money budgeted to pay for student travel and expenses, and attention paid to quality control and to ensure that helpful feedback is given to the project owners even if the project was not too good. Projects take a lot of student time and so students usually only do one. Consequently, projects trade off exposure to the real world against variety — the students come to learn a whole lot about one problem.

Another approach might be to have students read articles describing successful O.R. applications, for example the Interfaces articles from the finalists of the Edelman Prize competition. Reading these articles, however, gives very little sense of the grunt work that had to be done before the data needed to run the model could actually be used in the model, or why that particular model was chosen. These articles are about O.R. but omit much of the "analytics." Further, in our literature, we do not find articles describing how a senior decision-maker improved their intuitive decision-making as a result of being fed cleaner of more relevant data.

Most business schools could not devote the class time or resources required to have non-O.R. "majors" do a substantial student project. The Ivey School of Business is one of these. We have had a single integrated statistics/O.R. core course in our degree programs (BA, MBA and EMBA) for many years, which we teach using cases. Cases have some advantages over projects in that they can expose students to real-world problems without the administrative issues, the budget and concerns over quality control and feedback to clients. And you can do a lot of cases in a course: Ivey's undergraduate program covers just about one case per session for 30 sessions, so students see about 30 real-world problem situations where the use of O.R. or statistics can be beneficial. Cases have an advantage over Interfaces readings in that they do not (usually) include "answers," so the student has to try to solve the problem.

We teach O.R. using cases because we feel that there are major pedagogical benefits in terms of student involvement and active learning, but it occurs to me that cases might also be a good way to help students grasp some of the skills involved in doing pre-O.R. analytics. However, most of the cases we use are not as strong in the grunt work as they could be and perhaps ought to be. Harvard Business School cases are renowned for including a whole mass of extraneous material, so students reading the cases know that they have to dig out the kernels and discard a whole lot of chaff, but most O.R. cases are highly focused. It is not uncommon that every number that appears in the case is needed somewhere in the model! We could do a much better job at teaching pre-O.R. analytics if we ensured that the cases we use provide the student with the opportunity to do some grunt work before starting on the O.R. model. That said, here are some cases that we use regularly that do provide students with an opportunity to do some basic analytics:

• BriLux — The FOT-320 Decision: This R&D decision can be analyzed using a simulation model, but before the model can be constructed the decision structure must be teased out and, although information on the uncertainties is given, this must be interpreted and organized.
  
  
• Brent-Harbridge Developments: A decision analysis can be conducted to determine the best sequence of bids in some property auctions, but the probabilities of winning have to be extracted from historical bidding data (which usually requires several multivariable regressions).
  
  
• Northwest Newsprint: This quite long case includes a lot of detail about the company and includes many issues. Some effort is required to first focus on the supply chain optimization issue and then to formulate this as an optimization.
  
  
• Four Star Motorsports: Decisions on product prices suggest an optimization, but this requires that a set of demand model be developed from the available data.
  
  
• Dofasco Lance Desulphurizing Plant: This is a queueing problem, but the data is given as samples of raw arrival and service times. Use of a queueing model requires curve fitting and development of appropriate summary statistics.
  
  
• Lansink Appraisers: This case includes a regression model, but an understanding of the grunt work that goes into building a regression model leads to the realization that this is not a regression model that anyone should be using for anything serious.
  
  
• The Stephen B. Roman: This case describes a company with a problem, but it is not at all clear what problem should be addressed and what O.R. tools might be useful.

It does not seem to be a big step to include more "analytics" materials in our O.R. courses; it simply requires exposing our students to much more of the grunt work that most O.R. practitioners would see as an essential part of almost every O.R. project. We already have some tools to do this: projects, readings and cases. It would, however, be helpful if we had more cases that provide an opportunity to do more grunt work in advance of the O.R. modeling, and if more of our application readings included a richer background on the work that went on before the O.R. model was formulated. Of course, projects done in real organizations would be best, if you have the time and the resources to manage this effectively.



Peter C. Bell (pbell@ivey.uwo.ca) is a professor of management science and information systems at the Richard Ivey School of Business, University of Western Ontario.

Table of Contents

OR/MS Today Home Page


OR/MS Today copyright © 2008 by the Institute for Operations Research and the Management Sciences. All rights reserved.


Lionheart Publishing, Inc.
506 Roswell Rd., Suite 220, Marietta, GA 30060 USA
Phone: 770-431-0867 | Fax: 770-432-6969
E-mail: lpi@lionhrtpub.com
URL: http://www.lionhrtpub.com


Web Site © Copyright 2008 by Lionheart Publishing, Inc. All rights reserved.