Spent some time looking at DataXu today and MediaMath yesterday, both demand-side platforms for display advertising. Having spent so much time on search-related marketing and advertising, it’s good to look at the state-of-the-art in the display world. Stepping back though, my observation has been that the innovation is mostly around scale and performance as opposed to optimization.
Based on my preliminary look at DSP’s, here’s the recipe I’ve hypothesized: You use your favorite non-linear optimization algorithm e.g. machine learning and feed in as much data as you have available for the training set. Around that, you build integration with the major ad networks, exchanges, etc. and provide a UI that gives users control over a few levers on the optimization algo as well as campaign management and finally wrap client services around the full package and voila – you’re ready to build out your direct sales force, and that’s when the fun and the risk really ramp up.
Of course, those product steps are never as easy as the theory would suggest and answering thousands of bid request queries from the real-time bidding engines requires serious engineering.
I’m really looking forward to the next step in the evolution of marketing platforms – when the SEM and display advertising platforms merge, along with log-analysis/Omniture/Google Analytics-type infrastructure – to provide a truly compelling attribution and media mix optimization platform.