Today, Turn operates one of the largest marketing platforms on the Internet. According to comScore’s 2010 December Media Report
, Turn was ranked 6th in US audience reach, just behind companies like Google, Yahoo! and AOL. We’ve also won multiple awards, most recently being named by AlwaysOn for this year’s OnMedia Top 100 Company
list in the category known as “Technology Enablers.”
Whenever these industry benchmarks come up, I always get pinged with the question, “So how does the platform work? What’s in the secret sauce?” It’s a fair question, given that from the outside there’s almost no way to tell the different DSP vendors apart. While we can’t publicly talk about what the recipe to our secret sauce is, it’s still helpful to understand what some of the key ingredients are. Over the next two blog posts I’ll break down some of the key technology ingredients that help our clients meet their advertising goals, as well as work through the challenges facing data-driven advertising.
Data, Data, Data
Data enables you to understand an impression, and therefore its value to a campaign. At Turn, we use custom constructed servers, device drivers and proprietary in-house developed software for data storage. Our scalable infrastructure enables us to read an individual user’s data profile from among hundreds of millions of profiles within a very small time frame, generally 2 or 3 milliseconds. And, we do this over 100,000 times a second (8+ billion times a day).
Relevant data such as demographics, life stage, lifestyle and commercial intent is now widely available by credible vendors who protect the user’s privacy. However, our technology also enables advertisers to funnel their own data through the Turn platform. In 2011, CRM data will become a more vital part of the digital advertising mix, and a powerful component that enables marketers to leverage 3rd party data and combine it with data that they collect on their own consumers. It’s a simple equation where more data = smarter bidding, and combining CRM data with third party data gives the advertiser a significant competitive advantage for bidding on exchanges.
Machine Learning & Optimization
Machine learning is a sophisticated tool that Turn utilizes for our customers. Each machine learning model in our platform is custom for each advertiser, because the variables that matter for one advertiser differ from another. There is an old rhetorical question that goes something like this: if you know everything about a user, can you predict exactly what they are going to buy? The answer is yes, but only for an instant, since multiple variables about the user change in real-time.
Turn’s proprietary analytics methodology was developed by our staff of skilled PhDs and has multiple patents pending. It is leading edge technology because there’s a real need to score models in real-time. But what does ‘machine learned modeling’ really mean, and what role does it play? It’s technology that utilizes all of the data available to make highly accurate predictions. Predictions of click rates, predictions of conversation rates, and ultimately predictions of the CPM value for each impression available to the advertiser. This forms the basis of understanding what an advertiser CAN bid for an impression. An additional layer of ‘bid strategy’ algorithms are then applied to determine what the advertiser SHOULD bid for each impression. In ad exchanges, it’s important to adjust bidding strategy given what the other buyers in the market are doing. By combining predictive bid optimization with game theory-based bid strategies, Turn’s platform dramatically improves the media buying efficiency for both branding and direct response objectives.
In the next part of “The Ingredients of our Secret Sauce” we will walk through Turn’s scalability and ability to consistently achieve superior ROI. Stay tuned for more information!
VP of Engineering