We have taken on the management of 100% of the remnant video inventory for one of our clients, an award-winning video platform that connects content owners with advertisers. One segment of this inventory, including a group of 5 UK-based websites revolving around the football vertical, was showing some margin for improvement.

The incoming traffic was scattered all over the globe, with US (14%) and UK (12%) having the lion’s share, but with a relevant share of the traffic highly fragmented amongst other 110 countries.

After careful consideration of the bid landscape during the previous months we have decided to distribute the countries in 11 rules, grouping together countries with similar eCPMs in order to manage floorprices in the most effective way and command higher bids in every area.

The grouping was done following not only the eCPMs but also cultural, geographical and economic criterias. This strategy started to pay off pretty soon with a 13% increase in the overall eCPM on the first month and another 5% on top of that month after.

At this point, once we were happy with the level of yield of the sold impressions, we started to focus on the coverage as well. We have restructured the ad-stack including a video fallback system that reduces the amount of unfilled impressions, giving us an extra 5% on the amount of impressions sold, which in turn produced a 13% uplift revenue.

Next step has been to concentrate on driving up ad quality and revenue, making planned concessions on the coverage side in order to achieve more on the yield side. So, for example, we accepted a slight 3% drop in the overall fill rate because this secured a 11,7% uplift in the eCPM, month on month.

Given the turbulent nature of the supply in this vertical, with days featuring epic matches providing even 5 times the “normal” levels of traffic we were facing some difficulties when dealing with forescasts and mid-term strategy. In this field we can have huge variations depending on the day of the week, the time of the year, the matches being plated, and even some random events. This is why we decided to implement a specific module into our algorithms, customised on the fixtures calendar and taking into account a range of features of the match and the clubs involved to predict how popular a certain match will be and hence the levels of traffic we are likely to expect on our managed properties.