Beyond DCO: A Technical Whitepaper

Mar 09 2016

If you ever wanted to take a long look under the hood of high quality DCO, then you have come to the right place! We are regularly asked to explain why our Spongecell optimization is superior in form, speed, and function to traditional DCO. We have written a technical whitepaper on the topic for those who wish to go deep into the mechanics behind our solution and how it compares to more traditional methods of Dynamic Creative Optimization (DCO).

Spongecell employs two forms of creative optimization that improve both the impact and efficiencies of traditional A/B testing. Our objective when building this capability was to learn faster, optimize sooner, and continue learning more effectively. The core of Spongecell’s optimization solutions is based on an approach most commonly known as the “multi-armed bandit” model. (For more on how multi-armed bandit models work, download our whitepaper.)

Our first solution, Creative/Concept Targeting, uses weight-based bandit experiments and is closer in relation to the more traditional manual DCO model, although improved by mechanics and function to allow for faster evaluation periods for earlier optimization.

The second, affectionately known in-house as “The Brain,” externally referred to as Creative Auto-Optimization, was released as part of our CORE technology. Our always-on Creative Auto-Optimization makes further advances to predict how a consumer will react to a piece of creative content.

Our approach to Creative Auto-Optimization allows us to always be testing and learning based on audience cohort, and impression context, which in itself becomes a form of prediction to increase creative content relevance. This also provides the ability to optimize earlier and more efficiently than traditional A/B testing, and even our Creative/Concept Targeting.

Additionally, the Creative Auto-Optimization approach allows us to compare differences between different audience cohort models. A visual demonstration of this comparison will be released Q3 2016 in the form of “auto-discovery segments.” Auto Discovery segments will provide unexpected findings about how creatives perform under various contexts, outside of what humans could preconceive. Auto Discovery insights could be set as hard rules or recommendations on the campaign, or applied to future media and creative strategies.

Creative Auto-Optimization is preferable to run across all campaigns at all times in order to increase efficiencies, responses, and creative relevance. These efficiency gains are further magnified if the campaign has regular creative refreshes, has programmatic creative variations, has limited impressions, or a limited promotional period.

Creative Auto-Optimization does not have to replace your existing optimization; the two are complementary. Creative Auto-Optimization considers the impression-context uniquely, so previous or concurrent media optimization does not upset its function or result in any way. Impact and user experience will, in fact, be more favorable. Feel free to challenge us and take peek or a ponder at the mechanics under our hood. We welcome all DCO drag-races!

Need more? Download our Creative Auto-Optimization whitepaper.

By Byron Ellis, CTO

Byron Ellis is CTO of Spongecell where he heads research and development. Prior to Spongecell, Byron served as Chief Data Scientist for LivePerson, a company that helps businesses engage in real-time customer communications. He also authored Real-Time Analytics: Techniques To Analyze and Visualize Streaming Data where he draws on his experience building large-scale data processing centers and gives insight into how to build an effective real-time analytics platform from the ground up. Byron also held various leadership positions at AdBrite, eventually serving as the company’s CTO. Byron holds a PhD in Statistics from Harvard University and was a postdoctoral fellow at Stanford Medical School. Byron earned a BS in Cybernetics from UCLA.