Integration of Predictive Analytics into the Enterprise

Home/Blogs/Integration of Predictive Analytics into the Enterprise

Integration of Predictive Analytics into the Enterprise

A couple of decades ago I had the opportunity to work on a business problem that initially seemed rather daunting. The challenge was to optimize the operations of a precision engineering goods manufacturer by moving to a completely different approach to planning. Our company initially used to follow the practices that were derived from the management thinking that worked with the economic environment of the 70s, when the question of how to always have stock of whatever our customers could want was answered by always ensuring that lots of everything was always available. Somehow it always resulted in piles of inventory all through the manufacturing process and even then we frequently didn’t have what the customer wanted.

Tuning in to the future. And so, we moved to adopting a different perspective – one that was based on a set of Japanese principles pertaining to the identification and elimination of waste of all types. Fundamentally, it could be termed as a shift from a make to stock view to a make to order view. Instead of producing stock that was in economic lots, and in anticipation of future demand, the philosophy moved to letting production respond to demand efficiently, and tailoring itself to adjust cost in order to be economical at the customers’ level of demand. In other words, the future was made a bigger and more real factor in deciding what to do and how to function.

Today those same principles form the fundamentals of what is now called Lean Manufacturing, and when combined with the principles of Six Sigma I have seen that they can be truly transformational. They worked very well for us back then as well, except that our knowledge about it had to be learned differently because the term Lean Manufacturing wasn’t yet commonplace at the time. It was while implementing these concepts that I got to write several pieces of what eventually became an ERP system to provide better information and problem solving support to all the manual and visual techniques that were applied all through the manufacturing chain.

The need for better demand forecasting. The result of all the work was a system that implemented the new practices into enterprise IT, and included the integrated use of what are now called prescriptive analytics techniques (such as for shop floor scheduling optimization). But even so, there was a hole or a lacuna felt in the entire system at the enterprise level. Sure, the system was supposed to respond optimally and cost-effectively to any kind of customer demand, but then it soon became apparent that a demand forecast was still badly required. For this, of course, predictive analytics could be used to provide a solution.

A predictive model of demand is of course based on the use of past data, not only of production and sales, but of additional variables (very possibly involving Big Data) that serve as indicators about customer behaviour and the external environment. But even with good quality data, the issue is that the data is only reflective of the past. It is possible to tweak and smoothen the models to account for some known influences of the future, and this must be done because the past cannot be the only indicator of the future.

The limits of usefulness of the past. This is perhaps one of the best reasons to illustrate why a team of data scientists working with even the most advanced tools for predictive modelling cannot work without closely involving the right experts from the business. Data, when pulled from across many functions of an enterprise can be incredibly complex to interpret correctly, but even when this interpretation is available there will always be additional factors and events about the future that will be known to and best understood only by the business experts, and which cannot be revealed by a dry statistical extrapolation of the past. These could include information about market events, new regulation, economic changes, competitor action, new sales channels being launched, and so on.

The lesson here for me was that even with all the most sophisticated management techniques being built into IT systems in every function in the enterprise supply chain for the purpose of optimizing it the future can remain a blind spot that hampers decision making. If predictive analytics is used and integrated to work with the IT systems there is no substitute for the complete involvement of business experts to provide useful supportive guidance for important management decisions about the short to medium term future.

About the Author:

Mario brings more than 24 years of professional consulting experience across North America, the UK & Europe, the Mediterranean & Africa, Asia, Australasia and India. In the course of his career, he has grown and managed various types of consulting and service operations and organizations.