Forum: Has operations research outgrown operations research?

Many quantitative fields are converging in their aims and fundamentals to a common concept domain – analytics.

By Michael E. Beyer, CAP

Outgrowing OR

The term “operations research” (O.R.) [1] has always been backward pointing, indicating more where we came from than what we currently do. Born of wartime necessity, our discipline’s original charge was to literally perform research on military operations, with the hope of maximizing effectiveness – prime examples being Abraham Wald’s insight into where to reinforce a bomber’s armor or the work of Britain’s operational research teams in determining the optimal depth charge settings for sinking U-boats.

The striking success of O.R. during World War II led to its adoption for peacetime use, notably by areas most analogous to military operations: logistics, manufacturing, supply chains and risk analysis. The 1950s through the 1970s were a golden era for operations research theory and application, both within and outside of the military. It was also during this time that many universities created O.R. schools and departments, further establishing this term as the name of our discipline. There have now been several generations of operations researchers, as well as numerous textbooks and periodicals containing or referencing the term “operations research.” On the economic side, a quick search for the job title “operations research analyst” turns up page after page of employers seeking someone with such a background, with interested organizations ranging from military and federal offices to manufacturers and financial institutions. It is safe to say that operations research, as both a term and a field, is very well established.

While O.R. was becoming more “civilian,” other quantitative fields were also developing and evolving. In particular, management as a body of theory and practices was becoming an independent academic discipline in its own right, while economics was becoming increasingly mathematical, with an emphasis on using optimization as a driver for economic behavior. In addition, entirely new fields emerged related to the nascent computer industry. These independent strands would eventually lead to the modern fields of statistical quality control, econometrics, machine learning/data mining and data science. However, unlike O.R.’s military upbringing, these fields have political and scientific roots, leading to different cultures, terminology and areas of focus and application.

Why is this brief and very incomplete bit of history relevant? Because we are seeing something in the 21st century that was not apparent in the 20th: Many of the above fields are converging in their aims and in their underlying mathematical fundamentals to a common concept domain – Analytics. Analytics’ three sub-categories  – descriptive, predictive and prescriptive analytics – nicely capture the breadth of these various fields. At this point, there is a profusion of disciplines whose aim is to use data, mathematical modeling and computation to help improve organizational effectiveness. Where there used to be O.R., statistics and economics, there is now [insert-adjective-here] analytics, data mining, business intelligence, data science, machine learning, statistical learning, risk analysis, predictive modeling, industrial engineering, financial engineering, information engineering, etc., and the list goes on.

Each of the above fields has a legitimate claim to using advanced analytical methods to help organizations operate better, whether it be smarter risk-taking, better accuracy, more complete data, or a finer-grained understanding of the business or operational environment. So where does this leave O.R., or more pointedly, INFORMS? O.R. and INFORMS represent perhaps the most established thread of the analytics disciplines, with a well-developed identity and base of techniques. But, in light of the recent expansion of other fields into areas traditionally seen as the domain of O.R., perhaps it is time to reconsider what it means to be an O.R. practitioner vs. a data scientist or analytics professional. To this end, I’d like to offer a couple thoughts and my personal experience with being an “O.R. guy” who almost exclusively works with non-O.R. professionals. I will conclude with a couple of suggestions for the future direction of INFORMS and O.R.

First, we (i.e., INFORMS and the broader O.R. community) need to acknowledge that there are many professionals out there that use advanced mathematics and data to help improve operational effectiveness, apart from those who would ever identify as “O.R. practitioners.” These other disciplines utilize almost all of the same mathematical/analytical techniques that O.R. does: statistics, probability, mathematical programming/optimization and modeling. What differs the most is the culture, focus and terminology of each field.

Second, this situation has parallels in the not too distant past. In particular, there is an interesting historical analogy that captures many aspects of the relationship between O.R. and the broader analytics community: What is now “environmental engineering” used to be called “sanitary engineering.” The latter (and earlier) term indicated the initial focus on public health risks, notably water/wastewater engineering, as waterborne illnesses were a major issue in the late 19th and early 20th centuries. The term “sanitary engineering” persisted for some time, well into the era when many sanitary engineers were becoming involved in more than just sewage and public health. Eventually, sanitary engineering came to be seen as a subset of the modern field of “environmental engineering,” a field that today admits a wide array of practitioners with diverse backgrounds [2].

My personal experience with being an O.R. professional who works exclusively with non-O.R. colleagues has also taught me a few things about the power of terminology. To start, my career has pivoted several times. I started out working for an environmental engineering firm, then an environmental and health science consulting firm, and now I work with an IT consulting firm, creating data products related to biomedical research. Over my career, I have held the title “engineer,” “consultant” and now “data scientist.” Each new role has required me to “re-brand” myself and find new uses and ways to describe my O.R. skills. I think this personal experience in re-branding gives me a unique perspective on the current re-branding discussions surrounding INFORMS and O.R.

Here’s what I’ve experienced. When I ask people what comes to mind when I say, “I do operations research,” I get, “you study how things are done” or something to that effect. However, when I ask people what they think I can do when I say, “I do analytics,” they say, naturally, “analysis” (it’s right there in the root!). When I probe deeper, people indicate that they assume its “mathematical,” with “statistics and computers.” They also seem to agree that I would be someone they would talk to about the goals “optimize,” “analyze,” “describe” and “predict.” In other words, to the uninitiated, O.R. is too evocative of the study of how things are done and not with what should be done (prescriptive analytics). It also does not seem to convey the degree of analytical sophistication involved. As much as I like considering myself an “O.R. guy,” it has been my experience that the term “analytics” simply gets the point across a lot better than O.R. (and certainly for any “elevator speech-type” interaction).

The above points and my own experiences suggest that we cannot afford to be insular, continuing to view O.R. as it was in the 20th century – the dominant bellwether. The landscape has changed, and so we need to adapt as well. Either (a) we change the name of our field (and hence INFORMS) to something more naturally evocative (à la changing “sanitary engineering” to “environmental engineering”), or (b) we stay as INFORMS and O.R. but we work to create a more formalized mechanism for collaboration among the different disciplines in this space and for creating a shared vision of what we, as a group of related fields, are and (just as importantly) what we are not.

For (a), I would suggest we consider something like “Decision Analytics,” as it emphasizes the use of mathematics and data analysis to make better decisions (as opposed to predictions or inferences). It’s evocative of what we do and brings to the fore the key mindset that all O.R. practitioners possess and where we can add the most value. The details (e.g., decision analysis vs. simulation modeling vs. mathematical programming) are best left to one’s resume.

For (b), I could see a new “umbrella” organization being formed, akin to the International Federation of Operations Research Societies. This new “organization of organizations” would have the most legitimate claim to representing analytics as a whole (i.e., unqualified). Such an organization (say, the “Confederation of Analytics Societies” [CAS]) would help conceptually organize the landscape of analytics professions and societies, as well as coordinate joint conferences, publications/newsletters and outreach/branding of the analytics practice domain.

INFORMS is well placed to champion the creation of such an analytics umbrella organization (or even wholeheartedly become that organization). We have already gained first-mover advantage in analytics via Analytics magazine and the Certified Analytics Professional (CAP®) program, both of which help to define what it means to practice analytics. The CAP exam would be a great starting point for engaging member societies in cross-discipline collaboration, bringing together business intelligence, decision analysis, “big data,” data science, statistics and O.R. to help create a “CAP 2.0” that embodies the voices of the broader analytics community. Finally, CAP 2.0 would help address the “demarcation issue,” as even something as broadly defined as “analytics” cannot be all things to all people; thus, CAP 2.0 could help refine the scope and value proposition of analytics as a whole. Under the scenario presented above, INFORMS could very well stay INFORMS and O.R. stay O.R.; what would change would be the explicit acknowledgement and support of the idea that we are one voice among many in this new world of analytics, which requires a far more ecumenical posture towards O.R. and analytics than in the past.

Of course, O.R. is not the only field having identity issues. As some of you may be aware, our sister organization, the American Statistical Association, has been having its own internal struggle with how to define and market itself amid the rise of “big data,” machine learning and data science. Similar to O.R., statistics is a venerable but relatively staid field compared to the newer disciplines. These new fields, in attempting to define themselves, sometimes come across as impetuous or grandiose in their claims. Certainly, they have very valuable things to offer and are to be welcomed into the larger analytics community. However, with age comes maturity (at least for academic disciplines), and both O.R. and statistics are still just as relevant, if not more so, amidst the current wave of excitement surrounding large data sets and rapid data analysis. Statistics offers the inferential viewpoint, with its concerns about replicability, robustness and proper interpretation, whereas O.R. frames the key decision problems, identifies their relevance and analytical requirements, and creates mathematical models that allow advanced predictive and descriptive techniques to be leveraged into effective operational decisions.

Focusing the value proposition of O.R. on decisions is key. While we should clearly advertise our value, we should not appear to be taking more than our due. Analytics, writ large, encompasses far more than O.R.: It includes descriptive analytics (including estimation and inference), which rightfully belongs with the statisticians and data mining professionals, and predictive analytics, which is increasingly the purview of data science and machine learning. However, the area where O.R. really shines is in creating and optimizing mathematical models that relate data, inferences and predictions to decisions and the operation of the organization as a unified system. Thus, while good inferences and predictions certainly support good decisions, when one is talking about mathematical modeling of decisions qua decisions, then one is talking about O.R.

Therefore, the key question for INFORMS and O.R. is not if we should be broader, but how. Do we change our name and begin re-marketing as a new, more general brand, or do we keep our nomenclature and identity but actively and explicitly recognize, participate in, and support the existence of a broader analytics community. We should consider the tradeoffs involved: INFORMS and “operations research” have substantial cachet in the markets that would seek our services, so if we change our name and dilute our identity, we would be incurring a substantial opportunity cost. However, I also see the future of our profession as being one where we explicitly acknowledge our unique contributions to and membership in “big tent” analytics. On balance, I would rather see more effort on cross-pollination and the establishment of some sort of “umbrella” organization as I described earlier, as opposed to trying to re-tool INFORMS to compete with all the other analytics organizations out there; there are too many and they are too well funded and established.

In closing, we as O.R. practitioners have inherited the legacies of Dantzig, Howard and Blackett – advanced mathematics, a focus on decisions and their impact on entire systems, and the interdisciplinary spirit, respectively. We should be proud of this heritage and its enduring value to society while acknowledging the equally vital contributions from our sister disciplines as part of a broader analytics community. In deciding our way in this new landscape, we should heed the Greek maxim: “Know thyself”; therefore, let us take our place within the larger analytics community, aware of what we are, and, just as importantly, what we are not.

Michael E. Beyer (beyer1981@hotmail.com), CAP, is a principal data scientist at Net eSolutions Corporation, headquartered in McLean, Va. He holds a B.S. degree in operations research and an M.Eng degree in biological and environmental engineering from Cornell University. This article is an expanded version of a post made by the author to INFORMS Connect in May 2015.
References & Notes

  1. This article also considers management science (MS) to be roughly synonymous with O.R. For brevity, I have referred to this combined field simply as operations research – it is not intended to convey any indication of relative importance of O.R. vs. MS.
  2. See “Environmental Engineering – A Fable and a Challenge,” by Larry Gordon, director of the Environmental Services Division of the New Mexico Health and Social Services Department. It was published in The California Sanitarian, Vol. 7, No. 4, July-August 1969 and presented to the Southwestern Chapter of the New Mexico Society of Professional Engineers on the Feb. 22, 1969. This piece makes very interesting reading in light of the new arrivals to the “math, data and decisions” space traditionally occupied by O.R. – “plus ça change, plus c’est la même chose.” Accessed from http://hslic.unm.edu/resources/spc/docs/gordon/Environmental_Engineering-_A_Fable_and_a_Challenge.pdf on June 28.