Analytics Maturity

A young child begins by crawling, learns to walk, and eventually gains sufficient mastery to run. An analogous process is proposed by advocates of analytics maturity models: organizations must pass through three stages — descriptive, predictive, prescriptive — as they gain ‘analytics maturity’. (See for example the INFORMS Analytics Maturity Model and the TDWI Analytics Maturity Model.) Inherent in these models is the notion that ‘analytics’ is all about processing and analyzing data in a progression of increasingly sophisticated ways. When I see such ideas proposed, I find myself imagining the following discussion taking place in a corporate boardroom…

“Margret, I’m getting worried. We’re losing market share every month.”

“I know it Frank. Our supply chain is a mess, and our competitors are eating our lunch by offering much faster service. We need to optimize our supply chain.”

“That’s exactly what we need to do, but unfortunately, we’re just not mature enough! I’m afraid that by the time we are, we’ll be out of business. You know, it may be time for us to pull the cords on our golden parachutes! ha, ha, ha!”

Do you think such conversations are taking place? Does the notion of ‘analytics maturity’ make any sense? Let’s take a look.

Descriptive Analytics

We are in the realm of ‘big data’: unearthing insights from massive datasets using exotic software technologies such as HIVE, KAFKA, STORM and PIG. (Who names these things?) No one in the organization, except a few IT specialists, understands how this stuff works.

Predictive Analytics

Data scientists extract the maximum amount of information from datasets — often the same kind of datasets that people have been using for 50 years — to make predictions. They use advanced ‘machine learning’ and ‘deep learning’ methods, methods that no one else in the organization understands. Sometimes, even the data scientists themselves cannot explain how these methods work.

Prescriptive Analytics

Operations research analysts build mathematical models of real systems or processes in order to improve, optimize, or simulate them. Often the amount of data required is quite modest. These models, and the methods used to solve them, can be sophisticated, but since they are based on the relationships of entities existing in the real world, it is possible to explain — at least on a high level — how they work.

Do the above listed ‘stages’ represent the natural progression of steps in a single process? Should you wait years before you can use operations research to solve problems or improve operations?

Rather than using a model to determine your level of ‘analytics maturity’, let me offer you a simple test to determine whether you are ready for operations research:

If you are conversant with the concepts of more or less, and better or worse, you are ready.

1 thought on “Analytics Maturity

  1. Joe Q. Dangerously

    When progressing through and graduating each phase of analytics maturity, is it also required to progress from small data to big data? Where does that progression lie on the analytics maturity spectrum?

    For instance, if you have performed predictive analytics on small data sets and generated tens of spurious correlations as the basis for your predictions, are you allowed to progress to prescriptive analytics? Or must you cherry pick from the best of millions of spurious correlations on big data before being allowed to progress to the prescriptive phase? Or maybe the very act of cherry picking among the spurious correlations, which of course is optimization, therefore qualifies it as prescriptive analytics?

    P.S. It is amusing that “analytics” merits a squiggle in a reply to an article on analytics.

Comments are closed.