INFORMS Open Forum

Expand all | Collapse all

Can we improve the terminology of analytics?

  • 1.  Can we improve the terminology of analytics?

    Posted 11-29-2016 21:28
    My co-author and I currently are developing the 6th edition of our McGraw-Hill textbook, Introduction to Management Science: A Modeling and Case Studies Approach with Spreadsheets, and when that is done, I will be developing the 11th edition of the Hillier-Liberman textbook, Introduction to Operations Research. (My co-author for the former textbook is my son, Mark hillier, a faculty member in the University of Washington business school who keeps winning teaching awards for his management science and spreadsheet modeling courses.) I just have written a draft of a new section on the relationship between analytics and management science for the new edition of the former textbook (and later will write a similar section for for the next edition of the latter textbook). This draft uses the traditional terminology for the three phases of  the analytics approach (descriptive analytics, predictive analytics, and prescriptive analytics). However, my reaction when doing this was that, while this terminology works quite well for us "Insiders,"  its usage has some significant drawbacks when we reach out to students, managers, etc. It seems to me that we could do much better by using "plain English" names for these phases of the analytics approach.  I am writing to solicit feedback from others on this issue.

    I briefly describe below my objections to the traditional terminology, after which I will present my suggestions for improved terminology. 

    Phase 1: Descriptive analytics. Put yourselves in the shoes of an "outsider" and try to decipher what this name for phase 1 means. Analytical techniques that have steps that are well described? Analytical techniques that provide guidelines for describing something well? Or what? More importantly, it conveys nothing about what actually is to be done. The whole focus revolves around dealing with data, perhaps even massive amount amounts of data. It involves applying data science through the techniques of information technology for data warehousing, data mining, data cleaning, etc., in order to transfer the data into insight. However, the name for phase 1 does not even hint at the central role of dealing with data.

    Phase 2: Predictive analytics. This one isn't too bad. However, predictive isn't quite the right word. A dictionary definition of predict is "to state what one believes will happen," or "prophecy," as if one is just making an educated guess without any calculations about what will happen in the future. However, the actual activity involves forecasting, which involves performing calculations on historical data to obtain a statistical forecast of the future. Therefore, it would seem more appropriate to have forecasting in the name.

    Phase 3: Prescriptive analytics. A dictionary definition of prescriptive is "making strict requirements or rules." Managers certainly would object to analytics professionals imposing strict requirements or rules.  The focus instead is on using the techniques of analytics (i.e., the techniques of operations research in this case) to guide data-driven decision making. This ties in with the INFORMS definition of analytics as "the scientific process of transforming data into insight for making better decisions." Therefore, it would seem more appropriate to have decision in the name.

    With all this in mind, I list below my suggestions for the names of the three phases of analytics:

    Phase 1: Data analytics. This name already is being widely used and nicely conveys the focus of this phase.

    Phase 2: Forecasting-based analytics. This name nicely conveys the focus of this phase. 

    Phase 3: Decision analytics. This name nicely conveys the focus of this phase.

    For those of you who have been more heavily involved in the analytics movement than I have, I would greatly appreciate your reaction to the above. Should I go ahead with the above suggested names for the three phases of analytics? Or what is your advice? Textbooks can be quite influential and I want to make sure that I get it "right" in what I write about this important topic in the new editions of my textbooks. Thank you.

    -------------------------------
    Fred Hillier
    Professor of Operations Research, Emeritus
    Stanford University


  • 2.  RE: Can we improve the terminology of analytics?

    Posted 11-30-2016 04:19
    Edited by Marco Lübbecke 11-30-2016 04:19

    Just two quick remarks.

    1. the battles in Game of Thrones are nothing compared to the "interpretation wars" on the analytics notion, move forward with caution :)

    2. in Germany, where we usually speak German, the notion of eg. "predictive maintenance" (in German: "Predictive Maintenance") is something companies have a reasonable feeling about what that might be, some already use it. I believe this sounds much sexier to them than "forecasting" (which sounds like wheather forecast, which is never true anyways...)

    I currently conclude: this is through. The three notions are being pushed so heavily by big industry players that we probably should stick with them (until the next wave of notions arrives, that is...).

    All the best, Marco

    ------------------------------
    Marco Lübbecke
    Professor of Operations Research
    RWTH Aachen University
    Germany



  • 3.  RE: Can we improve the terminology of analytics?

    Posted 11-30-2016 08:33

    Dr. Hillier,

    Thank you for this topic. Just last week I requested access to your 10th edition text as a potential replacement text for my current text (Quantitative Methods for Business, Anderson/Sweeney...). After another review, it was not a good fit for my course in the MIS department in a College of Business. I have just reviewed your 5th edition IMS: A modeling and case studies approach with spreadsheets. It is a better fit and I will be requesting online access.

    Now to the point. I also struggle with these terms in the College of Business. As an OR/SA before retiring from the Army with my MS and PhD in OR, I find I am often bridging the gap between engineers and business/application. How do we describe what we are doing to the math averse? IMHO, here is my read on your 3 Phases:

    1. Data analytics may be the term of the decade, but descriptive analytics describes a broader field than just the data. As analysts we often need to describe the environment, limitations/assumptions/constraints, model design, EDA, etc. If you are addressing the statistics and visual/graphical representation of the data, then I feel Data analysis is appropriate. But it sounds more that you are setting the entry stages of analysis and thus descriptive analytics would be more appropriate.
    2. I agree that predictive analytics is more appropriate unless you are restricting yourself to forecasting. But, in doing so are you not eliminating other modeling such as regression? Although this does not predict the "future" as forecasting does, models do predict outcomes in relationships and can be used to predict the result of data not yet obtained such as what if we build a new engine?
    3. Whereas I recommend staying with your first two terms, here I support the change to decision analytics. (a) A point I and many analysts struggle with early on is that we present results to the decision makers but do not provide the answers. By using a term such as prescriptive it implies we are providing the answer. When a decision maker chooses another option than the "optimal" one we provided it would appear to violate the "prescriptive" result. (b) Secondly, this part of the analysis is in fact less prescriptive as it should address sensitivity analysis (relaxation of the rules) and providing of options to the decision maker. (c) Finally, I believe this is a great way to conclude and OR approach - preparation for the decision maker.

    Thanks again for this timely topic!

    Kaye

    ------------------------------
    =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
    President, UCA Faculty Senate, AY 2016-2017
    =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
    Kaye McKinzie, Ph.D.
    kmckinzie@uca.edu
    MIS, College of Business, 305 C
    University of Central Arkansas
    =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
    201 Donaghey Ave.
    Conway, AR 72035



  • 4.  RE: Can we improve the terminology of analytics?

    Posted 11-30-2016 15:15

    Fred, 

    I appreciate your efforts to find the best possible terminology for "Analytics".

    You are right, broadly used textbooks have profound impact of the subject area, and so the used terminology have to be chosen carefully.

    I agree that the "descriptive/predictive/prescriptive analytics" classification is confusing, and while they are supported by some major players, their interpretation managers, and even engineers outside of the CS/OR area, might have negative impact.

    I like the Phase (level) 1) Data Analytics, and Phase (level) 3) Decision analytics proposals, in my opinion these are excellent descriptive terms to focus on analyzing data and quantitative support for decisions.

    However, with Phase (level) 2) Predictive/Forecast-based  Analytics, my choice is Predictive Analytics. Not only "forecast" reminds me (as it reminds Marc) to weather forecast, but "forecast-based analytics" is a too cumbersome expression. Predict and Forecast are synonyms anyway,

    Best Regards

    Tamás 

    ------------------------------
    Tamás Terlaky
    Professor, Chair
    Lehigh University
    Bethlehem PA



  • 5.  RE: Can we improve the terminology of analytics?

    Posted 12-02-2016 02:31

    Dr. Hillier,

    I also appreciate your efforts to establish the best terminology to describe the different stages of analytics. The difficulty is selecting a single word to describe each of the stages. I have found the following to be useful when describing analytics (I would give the proper attribution, but have seen this in a few places and am not sure of the origin):

    • Descriptive analytics - what has happened?
    • Predictive analytics - what will happen?
    • Prescriptive analytics - what should I do?

    If given a vote, I would stay with descriptive and predictive, but switch from prescriptive to your recommendation of decision analytics. My concern with data analytics rather than descriptive analytics is that data analytics does not convey the meaning of "what has happened." I can analyze historical data, but I can also forecast data, optimize data, and make decisions using data.

    I do like decision analytics because I think it conveys "what should I do" much clearer than prescriptive analytics.  Also, as you may be aware, the term "Prescriptive Analytics," with both the "P" and the "A" capitalized, is a software product by Ayata and the name is trademarked.

    Best Regards,

    David

    ------------------------------
    David Hunt
    Oliver Wyman
    Princeton NJ



  • 6.  RE: Can we improve the terminology of analytics?

    Posted 12-02-2016 11:44

    Dr. Hillier,

    Exposing students to the descriptive/predictive/prescriptive taxonomy is highly desirable since that terminology is very prevalent in industry, thanks largely to efforts of large consulting organizations.  That said, you could present those terms and also explain, as you have done, why other terms may be more appropriate and why.  You could also subsequently use the more appropriate terminology in the remainder of the textbook.

    Terminology should be considered not only from the viewpoint of practitioners, but also from that of potential consumers of analytics.  As mentioned, large consulting organizations have educated potential clients using descriptive/predictive/prescriptive.  While these consumers have never really understood what Operations Research is, they at least think they know what analytics is in its various forms (thanks to Money Ball, etc.) and are spending freely in pursuit of analytics value.  Your textbook can be influential in providing practitioners with a proper understanding of the nature of analytics, but it would be a tall order to override prevalent industry terminology not only among practitioners, but also among those who consume and pay the bills for analytics.

    ------------------------------
    Erick Wikum




  • 7.  RE: Can we improve the terminology of analytics?

    Posted 12-03-2016 16:02

    Dr. Hillier,

    Difficulties with your suggestions include...

    1. Data Analytics -->  because data is used in all three phases, it's confusing to put the word "data" in the first phase as that wording would imply the other phases do not include data. Also, the common usage of "data analytics" has a meaning that applies to all three phases.

    2. Forecasting-based Analytics  -->  For many people, the word "forecasting" (to many people) brings to mind the techniques (e.g. exponential smoothing, regression) to create a point estimate. An important phase 2 (predictive analytics) includes stochastic simulation. Few people consider stochastic simulation as a subset of forecasting.

    3. Decision analytics --> decisions are made in all three phases. So it's misleading to include "decision" in only the third phase. Also, the expression "decision analytics" sounds strikingly similar to "decision analysis."

    I don't think any single adjective in front of the word "analytics" can clarify the phase well. Instead, as David Hunt suggests, a tag line is appropriate. A slight twist on his comments:

    1. Descriptive analytics - describing what has happened

    2. Predictive analytics - predicting what will or may happen  [observe "may" --> opens the door for simulation)

    3. Prescriptive analytics - prescribing how decisions should be made [a logical, rational approach such as optimization]

    While imperfect, as you noted, the above status quo nomenclature tends to be easy to remember, e.g. all phases ending in "ive" makes it easier to remember all three.  I haven't terminology that is a notable improvement upon the descriptive, predictive, prescriptive terms that already have some traction.

    ------------------------------
    John Milne
    Clarkson University



  • 8.  RE: Can we improve the terminology of analytics?

    Posted 12-03-2016 07:19

    I propose a very different decomposition, which cuts right to the essence of the matter:

    1. Font Analytics

    2. Figuring Out Analytics

    Font Analytics is concerned with such things as determining the optimal fonts to use for dashboards, making nice pretty displays (a.k.a. Visualization), putting on a good show, etc.  As a concrete example, In the world of missile defense, the rarefied top echelon of Font Analytics practitioners develop the giant flashing screens shown to visiting Congressmen/women, high ranking officers, and other dignitaries.  People controlling the purse strings like giant flashing screens - if it's giant and flashing, it must be good. The more giant flashing screens, the better.

    Figuring Out Analytics is concerned with figuring things out. .Figuring out what's happened, what's happening, what could or should happen, and what we could or should do about it.  Operations Research is a synonym for Figuring Out Analytics.

    ------------------------------
    Mark L. Stone
    Figuring Out Analytics Expert



  • 9.  RE: Can we improve the terminology of analytics?

    Posted 12-04-2016 11:03

    I second John Milne's opinion, particularly in regard to decision-making occurring at all levels. My only change would be to reduce "will or may" to just "may" in the description of predictive analytics. That not only opens the door to other methods besides forecasting (such as simulation), but perhaps helps mitigate "irrational exuberance" on the part of consumers (who might be prone to ascribing greater accuracy to statistical forecasts than is warranted).

    Paul

    ------------------------------
    Paul Rubin
    Professor Emeritus
    Michigan State University
    East Lansing MI



  • 10.  RE: Can we improve the terminology of analytics?

    Posted 12-12-2016 20:54
    I would like to thank the many people who responded to my posting of November 29, including sone who responded via private email.  I learned a lot in the process and the next editions of my textbooks will benefit from this feedback.  I would now like to present my conclusions and summarize what I will be saying on this topic in the next edition of our Introduction to Management Science textbook (currently under development) and my Introduction to Operations Research textbook (next in line for a new edition).

    I now understand better that the traditional terminology (descriptive analytics, predictive analytics, and prescriptive analytics) is deeply embedded in the analytics community and that analytics professionals (including consultants) are heavily invested in continuing to using this terminology with their clientele. Therefore, instead of using the new terms suggested in my posting, I now will be highlighting the traditional terms, but with the qualifications described below.

    Because of the limited clarity of the traditional terms, several academicians mentioned to me that they add a brief descriptor in parentheses (e.g., what happened?, what will happen?, what should I do?) when they introduce these terms. This helps, but I feel that it is important for greater clarity to also include information in these descriptors that identify the focus of the work being done in each phase (namely, analyzing data, using predictive models, and using optimization models). Therefore, I would suggest that others consider using the following terms and definitions that I now plan to use in my textbooks.

    Phase 1: Descriptive analytics (analyzing data to better describe what has been happening)

    Phase 2: Predictive analytics (using models to better predict what might happen in the future)

    Phase 3: Prescriptive analytics (using optimization models to improve managerial decision making)

    A few people actually liked my original suggestion to use data analytics as the name for Phase 1. During my discussion of this phase, I will mention that data analytics also is a widely used term for describing Phase 1 (and perhaps Phase 2 as well) and has the virtue of clarifying that the analysis of data is the focus of the work being done.

    Quite a few people liked my suggestion to use decision analytics as the name for Phase 3. During my discussion of this phase, I will mention that the name decision analytics is another way of describing this phase since it highlights that the purpose of this phase is to guide data-driven decision making.

    -----------------------------------------------------------------
    Fred Hillier
    Professor of Operations Research, Emeritus
    Stanford University







  • 11.  RE: Can we improve the terminology of analytics?

    Posted 12-13-2016 19:51

    Dear Fred,

    I've been following the discussion on INFORMS Open Forum
    and admire your great efforts to standardize terminology.
    I don't already know the three names that some people say are already ingrained.
    I'm writing now to comment quickly on your parenthetical expansions.
    Phase 1: Descriptive analytics (analyzing data to better describe what has been happening)
    Phase 2: Predictive analytics (using models to better predict what might happen in the future)
     
    Phase 3: Prescriptive analytics (using optimization models to improve managerial decision making)
     
    We often hear people saying "xxx will help us to better understand yyy",
    but I've always blanched at this turn of phrase, partly because of the split infinitive
    and partly because "better" is close to being redundant.
    These versions are a bit more brief and direct:   (I haven't changed the 3rd)
    Phase 1: Descriptive analytics (analyzing data to determine what has been happening)
    Phase 2: Predictive analytics (using models to predict what might happen in the future)
     
    Phase 3: Prescriptive analytics (using optimization models to improve managerial decision making)
    ------------------------------
    Michael Saunders
    MS&E and ICME
    Stanford University
    Stanford CA



  • 12.  RE: Can we improve the terminology of analytics?

    Posted 12-14-2016 09:07

    Dear Fred:

    I appreciate your effort to standardize the terminology of analytics. And, I like Michael Saunders' suggestions for the parenthetical expansions of your definitions. However, I have two suggested modifications of Michael's terminology. Both are in line with how I use models and analytics in my professional work as a policy analyst (providing decision support to public policymakers, and to private sector managers). Policy analysts (like most operations researchers) do not prescribe policies; they use a variety of tools (including descriptive analytics and predictive analytics) to improve decision making. They use models, but not necessarily optimization models. So, I would stick with Michael's Phase 1 and Phase 2 (note that he says 'models' are used in Phase 2, not 'optimization models'). But, I would call Phase 3  "Decision analytics", and I would define it as "using models to improve decision making".

    Warren Walker

    ------------------------------
    Warren Walker
    Professor of Policy Analysis
    Delft University of Technology
    Delft, the Netherlands



  • 13.  RE: Can we improve the terminology of analytics?

    Posted 12-14-2016 04:57

    Due to the embedding of "Descriptive, Predictive, Prescriptive" in our community discourse I've given up arguing for better-named categories or a more open-ended list (three is a magic number for closed lists).

    I continue to think that while this sequence ("Descriptive, Predictive, Prescriptive") can be reflected the pages of a textbook, there is no fixed requirement to address these kinds of analytics in sequence when it comes to application. As the business need drives the kind of analytics required, any selection or combination of techniques from these three categories can be used. Using "phase" (interpreted as phrases of analytics) can suggest a sequence of progression, so may I suggest using "category" (as a simple categorization of techniques) that carries no baggage of sequencing.

    Regards

    ------------------------------
    Rahul Saxena
    Cobot Systems
    Bangalore



  • 14.  RE: Can we improve the terminology of analytics?

    Posted 12-15-2016 12:54

    Fred,

    Interesting discussion, and I appreciate your efforts to poll the field.

    I heartily agree with keeping the descriptive, predictive, prescriptive paradigm, even with its shortcomings, given its adoption. Adding parenthetical statements is a good idea. But I have a concern about your addition for prescriptive:

    Phase 3: Prescriptive analytics (using optimization models to improve managerial decision making)
    I'd argue that all analytics, even descriptive statistics that are barely more than reporting, are done to "improve managerial decision making." I'd prefer something simple, although granted reductive, like "using optimization models to find the best course of action").
    Polly
    ------------------------------
    Polly Mitchell-Guthrie
    Sr. Mgr. Advanced Analytics Customer Liaison Group
    SAS Institute, Inc.
    Cary NC