Optimization is surfacing more frequently in the applications and tools markets. I was recently speaking with a vendor who competes in the content delivery network (CDN) market about their optimization technology. Consequently, I started thinking about optimization, its role in decision analytics, and how it is applied.
Optimization is one of nine techniques that I’ve identified that support decision analytics (see the ESG Market Summary Report, Decision Analytics: Building the Foundation for Predictive Intelligence and Beyond, August 2014.)
Optimization anchors the upper left hand corner of the decision analytics continuum that you see below.
Optimization represents the end-game for decisioning where resources are well defined and the objective is to minimize or maximize the use of resources. The utility of mathematical optimization is that it can determine the best way to leverage resources based on your objective. Most of us wish that the optimization story could end here – because this way an optimized solution would be simple, elegant, and immensely practical. However, the scope and frequency of optimization are critical to ensuring that the optimization is really optimal.
The scope of the optimization is important because unexpected events are always happening and having some flexibility to accommodate change without having to re-optimize is desirable. I remember a conversation I had with a senior airline pilot. He mentioned that every time a new CIO came on board, they would look for ways to increase resource utilization and tighten up their optimization criteria as a way to increase profitability. The resulting resource assignments typically did not have enough flexibility built in to accommodate small unexpected changes. Consequently, the airline was forced to re-optimize frequently which caused higher levels of churn of people and equipment which led to decreased profitability and employee satisfaction. This communicates the value of having an economic model built into the optimization or electing to loosen up the optimization criteria as ways to improve the practical flexibility of the optimization.
Optimization is also a point-in-time activity based on current resource availability and objectives. The challenge is that resource availability and objectives change over time. Also, once resources are committed to an objective it can be difficult, inefficient, or politically untenable to change direction even as new information and events unfold. This means that the frequency of optimization is virtually just as important as the scope of the optimization. So optimization is subject to the Goldilocks syndrome. Optimize too much and you generate churn. Optimize too little and you forgo opportunities. Therefore, you want to optimize with just the right frequency. The right frequency may be as simple as optimizing based on an event or it may be as complicated as incorporating predictive algorithms and economic models to determine when best to optimize.
Finally, I see many vendors using the term optimization to denote nothing more than an improvement over how they used to do something. While I’m not a fan of this practice, many marketing people are prone to a certain linguistic flexibility and consider definitions as mere guidelines. So the next time you hear the word optimize, and you’ll be hearing it quite a bit in the future, you’ll know just how overloaded a term it really is and that the devil is in the details when it comes to mathematical optimization.