Looking Under The Hood
JeanFrancoisPuget 2700028FGP Visits (3873)
Go to market (GTM) is something you must understand very quickly if you try to live from selling what you do. For optimization, we converged on a very simple view at IBM. There are two categories of customers, those who understand the technology, and those who don't. The formers are operations research (OR) practitioners and OR academics. The latter are line of business (LOB) people. Understanding how the technology work is not so much important to them. What is important is to understand what it enables.
The OR market is well served by IBM ILOG CPLEX Optimization Studio and its competitors, even if there is demand for better, faster, more comprehensive OR tools. The LOB market is usually served by packaged applications, eg supply chain management applications, or by the delivery of custom applications by consulting firms and IT integrators. These solutions (packaged or custom) shield LOB users from the mathematical models and mathematical algorithms used under the hood.
Is it that simple?
Can we really assume that LOB people do no need to see the math inside?
Some colleagues of mine (Mike Watson, Sara Lewis, Peter Cacioppi, and Jay Jayaraman) argue that there are very good reason to expose the math used under the hood to LOB users. Here are their arguments (full article is available here)
My colleagues state this in the context of the application (called LogicNet Plus) they are working on. It enables retailers and others to optimize the design of their supply chain distribution network.
I could stop here, but I do think that their view is valid in other areas where optimization has been used successfully in LOB applications. Let's look at each of their points in turn.
The first one is a quite generic one. The better you understand how the application works, the better use of it. Who would disagree with that? The real question to me is whether this can be reproduced in areas other than supply chain network design. Indeed, it may be that network design models are unusually simple. The good news here is that there are quite a few areas where we can explain the design of the mathematical model to LOB users. Well, that's my experience at least. Granted, there are also optimization problems with very complex mathematical model or where advanced techniques are used (eg column generation or benders decomposition). Those are not amenable to LOB users, and they must be used under the hood.
The second one is the most important one to me. This is something people don't necessarily have an intuition for: seemingly small changes in the business problem can lead to dramatic changes in how the corresponding mathematical model can be solved. Adding a tiny requirement here or there may lead to a much more difficult mathematical problem. The converse is also true, a seemingly large change in the business problem may be accommodated very smoothly by the mathematical model. Understanding some of the math helps build models that can be solved in a reasonable amount of time.
The last one is really about building confidence in optimization. This is also something that may not be granted, especially if the optimal solution looks counter intuitive.
Back to my colleagues, they explain the math inside their application in their Supply Chain Network Design book It is a must read for logistics people and more generally by people interested in optimization at work. You can also follow them on the IBM ILOG Optimization & Supply Chain Solutions Blog