Blogs

Analytics' best-kept secret is decision optimization

Product Manager – IBM Decision Optimization, IBM

In my undergraduate engineering studies, our class was given a competitive assignment to design the most efficient heat exchanger for a chemical production plant. Designing a good heat exchanger involves many conflicting decisions, such as selecting just the right input and output temperatures, the choice of materials, pipe diameters, pipe lengths and so forth. A tweak in the wrong direction can severely affect the system's efficiency.

http://www.ibmbigdatahub.com/sites/default/files/discoveranalyticsbestkeptsecret_blog.jpgConflicting decisions are also the case when designing an airplane, a production plant, a supply chain network or even a financial plan. To get the appropriate value, you need to decide on just the right combination of choices. So how does one go about finding the best combination of decisions for maximum efficiency?

One approach is simulation, which means creating a mathematical model of the system or business problem, and then trying—simulating—combinations of different parameters until finding the best combination. At the time, simulation was the best analytics tool available to our class, so we proceeded to spend an entire weekend trying different parameters and analyzing the results. Eventually, after 48 hours of nonstop simulation we reached the competition deadline and submitted our design.

Six months later, I joined the graduate engineering program at Carnegie Mellon University. There, I discovered the world of decision optimization under the guidance of Professor Ignacio Grossmann. Once again, our class was given an assignment to design the most efficient heat exchanger. But this time, we were given a new analytics tool I hadn’t heard of before—it was called mathematical optimization. Using this tool, we again created a mathematical model; however, instead of running hundreds of simulations, we specified an objective to maximize efficiency, clicked a Solve button, and five seconds later saw the best design.

I was hooked. How was it possible that mathematical optimization could do in five seconds what simulation took us 48 hours to do? Today, simulation and optimization are often used together, and for many decision problems optimization provides significant speedups and increased value.

Since that moment of realization in Professor Grossmann’s class, I’ve discovered the wonders of decision optimization technology across many domains, from beer manufacturing to oil exploration infrastructure investment to bandwidth auctions to financial planning to urban water systems design. The possibilities are endless.

Did our team win the competition mentioned previously? Yes, we did, but we could have found an even better design in five seconds. Had we known about analytics’ best-kept secret—decision optimization—we wouldn’t have had to miss out on an entire weekend of student fun.

Learn more from Susara’s theCube interview during the recent IBM Vision 2015 event.