Realizing Yieldex's Ten-Year Quest for Near-Perfect Delivery
Ever since the invention of the ad server in 1995, publishers have been working to optimize yield. At first, that meant selling as many impressions as possible. Soon, more sophisticated targeting meant that yield optimization and pricing became important skills. Then, with the introduction of real-time bidding (RTB,) the problem of allocation grew even more complex. But across the decades the objective has remained constant: fulfill commitments to buyers while maximizing revenue. As the co-founder of Yieldex, and before that NetGravity, I’ve been working on optimal yield delivery for over two decades. I’m incredibly excited to see it realized through the integration of Yieldex into AppNexus’ ad server, and delivering real-world yield increases at one of our largest customers.
Back in the early days of NetGravity, we struggled to solve one of the most difficult tasks in yield management: forecasting delivery to fulfill guarantees to buyers. Overlapping segments, buys of different sizes for different time periods, and fluctuating traffic all made forecasting horribly inaccurate. It was over a decade later when my Yieldex co-founder Doug Cosman patented a combination of clever algorithms to enable dramatically more accurate forecasting. This is how we started Yieldex.
From conception, Yieldex focused on yield optimization for publishers - as you can tell from the name we chose. Having built ad servers in the past, we knew that accurate forecasting could help drive yield both before the sale (by informing sales of availability), and after the sale (by informing the ad server of the most efficient way to deliver). So, we offered both – availability lookups for sales and delivery control for ad servers.
As the technology developed, we realized we had another incredible advantage over systems that weren’t forecast-aware: we had the ability to do a much better job allowing RTB to compete with guaranteed inventory. For example: typical ad servers serve guaranteed flights evenly throughout the day, meaning they try to deliver the same number of impressions during the 3 AM hour as the 3 PM hour. If RTB demand is roughly flat throughout the day as well, then at 3 AM (when there are fewer visitors) there will be a lot more contention than at 3 PM (when impressions are higher). A forecast-aware system would deliver more guaranteed impressions during high-traffic periods, capture more RTB demand during lower-traffic times, and deliver higher overall RTB revenue. We quickly patented this idea as well.
A startup’s journey is rarely linear. After months of work, we realized that the ad servers of the time (2008) simply didn’t have the ability to become forecast-aware – the APIs did not exist. Short of writing our own ad server, we didn’t know how to deliver on our promise of automated yield management, so we pivoted toward providing tools for ad operations, yield managers, and sales, enabling yield optimization through control of avails, pricing, and prioritization. Fortunately, we quickly found product-market fit with many of the largest publishers in the world. Preventing over-selling (and highlighting under-selling) with accurate forecasts, coupled with deep inventory analytics, proved to be invaluable to increasing yield.
One of the reasons we were so excited to partner with AppNexus in 2015 was the potential to realize our original vision – marrying our accurate forecasting with their state-of-the-art ad server to deliver higher yield. We began with basic avails integration, and then as we gained confidence, integrated progressively deeper. This year we achieved a never-before-seen direct integration between the forecasting and delivery systems. We engaged one of our largest customers, Schibsted Media Group, to help develop and test the forecast-aware delivery. Schibsted were invaluable in helping us look at the real-world outcomes and validating the efficiency increases.
We were thrilled to see substantial increases in programmatic revenue – just as we had predicted were possible nearly a decade prior. And, a much larger percentage of guaranteed orders delivered evenly all the way to the end of the flight, without manual intervention.
The production of Forecast-Shaped Pacing is the culmination of over ten years of work by the Yieldex team, in partnership with the AppNexus team, to deliver a key advancement in delivery optimization. Whether you call it artificial intelligence or just algorithmic optimization, our platform is taking over more of the complex repetitive work of yield optimization, freeing our customers for more strategic thinking.