This was posted originally in the Millennial Media tech blog.
The longer products are used before end-of-life, the higher the return on investment. Therefore, we plan our products to be robust enough to withstand market changes and technology evolution. However, in the ever-evolving area of campaign optimization at Millennial Media we build features to be used, learned from, and replaced quickly. Therefore, when we replace a feature, it’s more likely to be because of the feature’s success than due to an inadequacy. Quick end-of-life means that the feature was widely adopted and that we gained meaningful insights from its use, enough to justify investing in the next feature iteration. Paraphrasing Porras’ and Collin’s famous business book, we “build not to last”.
Here’s a typical example of how it may occur: Our data-science team comes up with an algorithm to partially automate a certain optimization process. Next, we build an interface to easily initiate, monitor and control the settings of that algorithm within each campaign. Our optimizers and account managers try it out, measure results and over time develop best practices. Once we have firm best practices we incorporate them into the algorithm. The partially automated algorithm becomes fully automated and the interface built to control it is rendered useless and removed. As soon as things “stabilize”, we start working on the next innovation that will construct and destruct once again.
While our features are in constant flux, the platform as a whole is intended to last longer. In fact, it’s the platform’s service-oriented architecture, in which independent components connect through APIs that gives us flexibility to iterate quickly. This is because services are de-coupled from each other and we can replace a given service without changing its connected services.
So the next time you hear that we killed an optimization feature, come congratulate the engineering, account management, and product teams involved for a job well done.
Comments
Post a Comment