Yesterday, INFORMS announced its voting pool for Official Definition of Analytics. The definition is as follows:

"Analytics -- the scientific process of transforming data into insight for making better decisions."

Although this definition of Analytics is similar to my understanding of Analytics, it makes me feel that Analytics is something for people mainly working with huge data like statisticians and computer scientists (capable of performing regression and time series analysis, multi-variate statistics, data mining, and information retrieval), and not someone who's working mainly with optimization models.

I'd would really appreciate it if you share your thoughts on this. In particular:

  1. Why do you think OR professionals should go towards Analytics, besides necessity of getting the next project? ;)

  2. What characteristics of an OR professional, other than knowing how to make better decisions, makes him/her a good candidate for the Analytics?

asked 09 Jun '12, 15:07

Ehsan's gravatar image

Ehsan ♦
accept rate: 16%

Your questions remind me of my post Is Optimization Part Of Analytics?

The proposed definition would lead to a yes answer if one focuses on the "making better decision" part. But the real question to me is: do we need to get insight from data analysis before making better decisions?

Sometimes the answer is yes. For instance one needs sales forecast in order to plan production. Such forecast comes from insight derived from data about past sales.

But there are also cases where optimization does not require insight derived from data analysis. For instance in fleet assignment, you are given a flight schedule, and you need to assign a physical plane to each scheduled flight. There is no insight derived from data here.

Back to your questions, answer to 1 is to prepare for the cases where optimization uses data analysis insight as input, as explained above.

Answer to 2 is that OR professional often have mathematical background useful to learn data analysis techniques.


answered 09 Jun '12, 17:53

jfpuget's gravatar image

accept rate: 7%

Dear Jean: Thanks. You answer to my first question is valid. However my concern is this is exactly what OR currently doing. In many OR project for many years, OR practitioners were using huge amount of data without calling it Analytics. As OR is a multi-disciplinary approach, OR teams usually consists of optimization, stats, and IT guys, depending on the problem scope/complexity. I suspect this re-branding (while good in nature as we're telling people what we have been doing for about 80 years is vital to what you think need now) would take the focus from the optimization aspect.

(10 Jun '12, 12:58) Ehsan ♦

I think there is fundamental reason for the shift from optimization to data analytics.

Optimization is very often used to reduce cost of operating something, through better use of resources.

Data analytics is often used to generate more revenue, eg analytics that models purchase patterns helping to sell more, or analytics that models churn for telco companies, helping them retain customers.

Add this to the fact that it is more interesting for a company to increase revenue than to reduce cost, and you're done.

(14 Jun '12, 15:46) jfpuget

I agree. Although this is changing in recent years as OR researchers and professionals in many fields such as SCM are starting to develop models for maximizing profits. However, some of the relatively old fields in OR such as revenue management and portfolio optimization are dealing with profit-maximizing objectives.

(14 Jun '12, 22:33) Ehsan ♦

Right. For instance, we have more and more applications in pricing optimization and yield management. These require a combination of predicitve analytics and optimization.

(15 Jun '12, 03:41) jfpuget

In my opinion, you can always get insight from the data even in "the fleet assignment".

Hardest part of the process is to develop the right model for the problem and you need to make some assumptions for the model. These assumptions may be hidden even for the customer. You can always relax some constraints or add some assumptions by just looking at the data.

(21 Jun '12, 04:34) Arman

Yes, you can always get insight. Question is: do you really need such insight to derive useufl decisions? Not necessarily in my experience.

What you need is a good understanding of the real business problem. This understanding can come from data analysis, but it can also come from the knowledge and experience of people dealing with the business problem. Extracting their knowledge as constraints and objective is often more valuable than data analysis.

Data analysis is useful for business problems that aren't well understood enough.

(21 Jun '12, 07:03) jfpuget
showing 5 of 6 show 1 more comments

A few years ago SAS were concerned that the term Business Intelligence (BI) was getting all the attention. BI is basically just reporting (descriptive statistics), and is done reasonably well be lots of software platforms. To distinguish their strengths they started using the term Business Analytics at every opportunity, and introduced the following graph, which I really like.

alt text

Basically, it separates Business Analytics into several categories. These are real distinctions, and SAS sells different software for each, but are somewhat artificial from a practitioners point of view since multiple types are often needed to deliver a desired result.

The idea is that as an organization's analytical capability matures they are able to move past BI, which describes the past, to more advanced analytics that help decide what to do in the future.

This doesn't answer your questions specifically, but I think it's relevant to the discussion.

Note that SAS keeps pushing the term Business Analytics, since in recent years the term "analytics" has been co-opted by the web-development community to mean website statistics.


answered 09 Jun '12, 19:57

DC%20Woods's gravatar image

DC Woods ♦
accept rate: 5%


I'll buy into any graph that asserts that we optimizers have a higher degree of intelligence than those stats guys. :-)

(10 Jun '12, 09:03) Paul Rubin ♦

Haha, yes that's exactly the reason I started showing it to people. As the OR guy at my company I used to constantly remind the lowly predictive modellers of the superiority of optimization.

(10 Jun '12, 09:15) DC Woods ♦

Thanks @DC. The graph is great. I will definitely use it as part of my future presentations to show off in front of stats and IT guys in our department. ;)

(10 Jun '12, 12:36) Ehsan ♦

IBM has been pushing a three-tiered model (think pyramid) for analytics that seems to resonate with many people. I'm paraphrasing from (suspect) memory, so apologies to IBM if I screw this up. The bottom tier is "descriptive analytics" (what the heck is going on here?). The middle tier is "predictive analytics" (what is going to happen if I do X? what is going to happen in the absence of any action/intervention?) The top tier, "prescriptive analytics", is where we optimizers live (now that I've got my descriptions and predictions, what the heck should I do?). In the context of the proposed INFORMS definition, optimization would be lurking in the phrase "better decisions".

Regarding your first question, I'm not sure we all need to stampede toward analytics, or at least not toward data molesting. (I'm too old to change.) That said, there is a growing demand for data analysis (especially large and/or real-time data sources). To some extent, the data side is the new frontier. Meanwhile, optimization is perhaps a bit more mature (with apologies to statisticians who've been running regressions since Gauss got out of diapers), and is finding its way more and more into canned software; so the need for more optimization experts may not be as great.

For the second question, I would hazard a guess that, besides technical chops, what my org behavior colleagues refer to as "openness to experience" might be important. If you're the sort that views everything through the same prism, or always wants to use the same tools, you may not be apt at finding information in sheep entrails. I tend to think that data dredging requires both an active imagination and a willingness to try this, try that, ...


answered 09 Jun '12, 18:05

Paul%20Rubin's gravatar image

Paul Rubin ♦
accept rate: 19%


One senior person at IBM had one more level: "Vindictive Analytics". Between descriptive and predictive, Vindictive Analytics concentrates on "Who messed up". A nice level to analyze.

(09 Jun '12, 22:57) Michael Trick ♦♦

I've seen that intermediate level referred to somewhat less cynically as "forensic analytics." Descriptive analytics focuses on what happened, forensic on why, predictive on what will happen, prescriptive on how we should act/react.

(Oops, didn't see the same comment in an answer below.)

(09 Feb '14, 11:50) Matthew Salt... ♦

from the standpoint of a professional organization, it makes perfect sense to give "operations research" and "management science" a meaning that is better accessible to the general public and to a larger number of well-trained persons. Hopefully, more people will understand what OR is about and why we need more not less.

as an operations researcher it makes me sad, of course, that this "new name" somewhat implies that operations research ever was anything less than giving perfect meaning to a wealth of data. Of course, since the very beginning this implied to help make better decisions.

as an applied mathematician (and @Paul, I am not yet "apostate", even though at a business school now!), I am afraid, of course, that "hard core" OR will disappear from INFORMS' radar. many OR/Analytics/MS/whatever problems are very, very difficult and complex, and addressing a lot more people (and nothing else happens by the name change) necessarily implies going "softer". A bad trend in my view.


answered 09 Jun '12, 19:43

Marco%20Luebbecke's gravatar image

Marco Luebbecke ♦
accept rate: 16%


Trust me, as soon as you joined a B-school, you became apostate in the eyes of "pure" mathematicians. :-)

(10 Jun '12, 09:05) Paul Rubin ♦

true. but I did not come unexpected. in view of the pure mathematicians, an applied mathematician is apostate by definition... so often one has problems being acknowledged properly in maths departments. but you know what? at a business school, it is - all of a sudden - exactly the other way round ;-)

(10 Jun '12, 09:19) Marco Luebbecke ♦

@Marco: I agree with you. The main concern would be to keep a good balance between optimization and data analytics approaches in academic environments as well as professional societies. This would become a main concern if it starts affecting OR people and curricula as some kind of management fad.

(10 Jun '12, 12:44) Ehsan ♦

Hi All,

First of all, thanks for bringing about this wonderful discussion on a great overarching field called Analytics. I tend to agree with the 3-tier pyramid definition (purportedly given by IBM-any reference?), where all of Statistics, OR and Computer Science (read Machine Intelligence/learning, Soft Computing/Computational Intelligence) would fit in quite comfortably. Having been qualified in all the three fields, I tend to advice the Analytics fraternity not to be jingoistic/nationalistic about their own field (Stats or OR or CS),as each of the three disciplines is subsumed by the word analytics and modern research frequently suggests the use of them in combination rather than in isolation in order to solve the problems better.

Unfortunately, the word analytics has become most 'abused' word nowadays, as every Tom, Dick and Harry talks about without actually knowing its true meaning.

My two cents, Dr Ravi Vadlamani


answered 07 Feb '14, 10:16

Dr%20Ravi's gravatar image

Dr Ravi
accept rate: 0%

edited 07 Feb '14, 10:21


The IBM view has been expressed in various ways, here is one in Analytics magazine

(07 Feb '14, 12:19) jfpuget

Re: Michael Trick's post, Vindictive Analytics is the vindictive cousin of Forensic Analytics.


answered 27 Jan '14, 15:29

Mark%20L%20Stone's gravatar image

Mark L Stone
accept rate: 0%

Re: jfpuget's link to , which says

Because of the computational requirements to solve an optimization problem, optimization is not applied in high-volume transactional applications.

Depending on what is meant by "high-volume" and "transactional" applications, nonlinear, even non-convex, optimization can and is used to solve large numbers of problems in real-time (not to mention oodles of LPs). Perhaps it comes down to what the value of solving those real-time optimization problems is.

I could tell you the details, but then I'd have to kill you.


answered 07 Feb '14, 13:58

Mark%20L%20Stone's gravatar image

Mark L Stone
accept rate: 0%

edited 07 Feb '14, 13:58

I'm not an author of the paper I cite, hence I feel free to not agree with all of it. In particular I know of transactional applications where optimization is used.

I agree with you therefore.

(07 Feb '14, 14:07) jfpuget

Also, optimization doesn't rule out the use of heuristics that can find good solutions in timely fashion, even for real-time applications.

(09 Feb '14, 11:54) Matthew Salt... ♦
Your answer
toggle preview

Follow this question

By Email:

Once you sign in you will be able to subscribe for any updates here



Answers and Comments

Markdown Basics

  • *italic* or _italic_
  • **bold** or __bold__
  • link:[text]( "Title")
  • image?![alt text](/path/img.jpg "Title")
  • numbered list: 1. Foo 2. Bar
  • to add a line break simply add two spaces to where you would like the new line to be.
  • basic HTML tags are also supported



Asked: 09 Jun '12, 15:07

Seen: 9,832 times

Last updated: 09 Feb '14, 11:54

OR-Exchange! Your site for questions, answers, and announcements about operations research.