Nice Graphics Always Win
JeanFrancoisPuget 2700028FGP Visits (3539)
I remember the first demo I made.
It was last century, in 1990 if I remember well. No, I wasn't using punch cards. Believe it or not, but at that time there were computers with a mouse, a windowing system, decent programming languages, and a good OS called Unix. Yes, this old fashioned ancestor of Linux and Mac OS X. Even Internet existed back in those old days...
Anyway, here was I, so proud that my solver was able to solve the zebra puzzle. The demo went as follows. I press return to run my executable, and a fraction of second later the result was displayed:
That's it, the output was a mere number, saying that the Zebra was in the last house, house number 5. To me it meant a lot, it meant that the solver was producing a correct answer to a non obvious problem.
I was really proud, so proud I couldn't resist showing this technical breakthrough to my colleagues, to my boss, whoever happened to be in the office. I was expecting applauses, really.
That didn't happen. The shock was immense, devastating. Each person I was showing it to wasn't impressed at all. Some were even wondering why I had been paid for several months just for replacing "hello world" by "5"!
That was my first lesson on the importance of nice result display. The second lesson came soon after, when we first tried to sell my solver. The prospect was a rail company that had a locomotive assignment problem. They wanted to assign locomotives to scheduled trains, while minimizing the number of locomotives. This is reminiscent of the fleet assignment problem in airlines for instance. I won't go into a detailed description of the problem here, but let me say we were competing with an established player. My solver was working fine, as fast as the other software to my surprise. But that wasn't enough for the customer, and they were leaning towards our competitor. Then I decided to use a brand new graphical interface tool that ILOG was developing. It displayed the results in a way very similar to what they were using. The minute I showed their problem and the corresponding result using their graphics I knew we had won. I could hear whispers, and I could see faces: people were recognizing their problem, and the results were meaningful to them.
The lesson was simple: optimization is not worth it if the results aren't understood by the business users. And the best way to make them understandable was to use nice graphics.
But there was a more subtle and important lesson. It was not optimization per se that mattered, it was the full process that went from customer data to business results. Optimization was only useful if it led to decisions meaningful for business. What mattered was prescriptive analytics, as Nathan Brixius defines it in his recent blog entry:
Prescriptive analytics concerns an entire process, and this process typically involves assembling data, building models, evaluating them, and presenting the results. Optimization comes into the play for the middle two steps, but the first and last are every bit as important.
Value is created by a process that involves more than calling a mathematical solver like CPLEX. Granted, without a robust and efficient solver the whole process is doomed. But the solver isn't all either. Then, as a software vendor it seemed quite interesting to provide tools that address each and every step of the process.
This is why IBM ILOG Optimization team is also working on tools to assemble data, tools to hook optimization to a business data model, tools to graphically display results, tools to deploy optimization in an IT environment, for instance in client server applications. These tools are bundled in our Optimization Decision Management (ODM) Enterprise offering. You can get more information about ODM Enterprise here.
The point is not that ODME is the only tool or the best tool around (well, if you really want to know we can talk... ) The point is that what matters is to cover all the business process that leads to useful decisions. We keep getting feedback that this is what really matters, at least when you try to make a living out of selling optimization software.