Analytics will not save OR/MS

Twenty years ago, I knew that I was being provocative when I submitted “Spreadsheets will NOT save OR/MS!” [1] to OR/MS Today, but I did not know how controversial it would be. Between the letters to the editor and the fact that it was still being cited 11 years later [2], I evidently struck a nerve.

But was I right? Looking back, I would say “yes” and “no.”

No, I was wrong. The importance of spreadsheet skills in business education combined with the willingness of OR/MS faculty to teach spreadsheets was a powerful combination. In some cases, it may have literally saved OR/MS courses in a business school’s core. In other cases, it made the material more palatable to students and served as a recruiting tool to get more of them into OR/MS electives. Maybe spreadsheets did save OR/MS.

Yes, I was right. Our courses survived, but what were we teaching? Other departments were happy to let us teach spreadsheet skills so that they could teach their content unchanged. We, on the other hand, had to figure out what OR/MS content to set aside to make room for spreadsheets. Our students, and colleagues, didn’t care about “the science of better.” They just wanted us to do their technical training.

Furthermore, a cursory look at textbooks shows that we often replaced the “algebraic curtain” with equally rigid spreadsheet formulations. From an intuitive standpoint, is SUMPRODUCT($B$3:$D$3,B7:D7) really an improvement over z = c₁x₁ + c₂x₂ + c₃x₃? 

New Savior

Now we have a new savior: analytics. INFORMS has enthusiastically embraced the descriptive, predictive, prescriptive analytics framework. Some would say that we were doing analytics before analytics was cool. But if what we were doing before “analytics was cool” was so valuable, then why did we need to jump on the analytics bandwagon to get attention?

Disclaimer: If spreadsheets and analytics are wrong, then I’m guilty. I have used spreadsheets as the primary software tool in my classes for more than 20 years, and I changed the name of my “Introduction to Management Science” course to “Introduction to Business Analytics” (and watched my enrollments double).

I’m not suggesting that we ignore opportunities for our classes to reach a wider audience. I’m just cautioning us to be careful how we do it and to be aware of the risks.

Just as our infatuation with spreadsheets put our content at risk, our embrace of analytics does the same. Those of us on the inside think we have a clear analytics framework [3].

Descriptive analytics

• Prepares and analyzes historical data.
• Identifies patterns from samples for reporting trends.

Predictive analytics

• Predicts future probabilities and trends.
• Finds relationships in data that may not be readily apparent with descriptive analysis.

Prescriptive analytics

• Evaluates and determines new ways to operate.
• Targets business objectives.
• Balances all constraints. 

However, here’s what I believe outsiders hear:

• Descriptive analytics → statistics
• Predictive analytics → more statistics
• Prescriptive analytics → blah blah blah. Some other stuff that I can ignore if I’m really good at statistics. 

Therefore, analytics = big data/data science/statistics and traditional topics such as optimization and decision analysis are, at best, relegated to minor roles.

How does this manifest?

• An accounting colleague told me that he encourages his students to take my analytics class because so many accounting ads mention big data. I’m happy to have his students, but he’s sending them to me without actually knowing what I’m teaching them.
• A marketing colleague wanted to add “analytics” to a course title because his class spent one week analyzing data with PivotTables. He did not understand how that was substantively different from the content in my analytics course.
• We had a proposed analytics major get sidetracked because we already have a data science minor and “those are pretty much the same thing.” 

It also impacts textbooks. I used one of the first analytics textbooks published. After using that text for about a year, I reviewed a proposal from a competing publisher for a new text that specifically criticized the former for too much “traditional operations research” content and insufficient coverage of data specific topics. Out of 14 total chapters in the proposal, the prescriptive analytics section had only two: one on spreadsheet modeling and one on linear optimization.

Feel free to argue amongst yourselves whether this is one or two chapters of “traditional” OR/MS content. I’ll recuse myself from that discussion while the 1999 and 2019 versions of me argue with each other.

Different Framework for Teaching Analytics

Instead of mourning the good old days or arguing about what OR/MS is or isn’t, I merely propose that we use a different framework for teaching analytics:

• Data-based, or data-first, analytics: This type of analytics starts with a data set. While mathematical theory may be needed for the process, no analysis can be done without data. Ex.: regression analysis, text mining, neural networks.
• Model-based, or model-first, analytics: This type of analytics starts with modeling relationships or decisions. There may be data behind the coefficients, but the data isn’t the focus of the process. Ex: linear programming, decision trees, simulation.

What are the advantages of this framework? First, it’s simpler for students. Second, it provides pedagogical balance between traditional tools and “data analytics.” Finally, it acknowledges that data and modeling are not mutually exclusive but are different starting points for looking at the analytics world and approaching problems.

Regardless of the text that I’m using, this is the framework that I use in my class. I explain the descriptive → predictive → prescriptive framework as a “practice” framework and point out places in our broader curriculum where various components of the framework are addressed, but I keep returning to data vs. model as the framework for the content in my course.

Based on recent reviews of student portfolios, it seems to be working and I welcome feedback on how others frame analytics for students in their introductory courses.

References

1. Groleau, T. G., 1999, “Spreadsheets will not save OR/MS,” OR/MS Today, Vol. 26, No. 1, p. 6.
2. Sodhi, M. S., and Tang, C. S., 2010, “A Long View of Research and Practice in Operations Research and Management Science” (Conclusion, pp. 275-297), Boston: Springer.
3. Retrieved from https://www.informs.org/Community/Analytics.