In a short time, header bidding has changed the industry for all parties in the programmatic supply chain. No longer is a publisher’s inventory available from just one or two separate exchanges. We now exist in a world where publishers are running multiple header bidding wrappers and within each wrapper, they may have four or more different exchange partners integrated. To add further complexity, each exchange is now receiving every impression on the page, not just what was left unsold.
What does this mean and why does it matter? There has been an explosion in the number of impressions and bid requests that need to be processed by SSPs and DSPs, alike. Processing this volume of impressions is costly for all parties, thus SSPs and DSPs need to intelligently manage the influx. At PubMatic, we are focused on supply path optimization, providing the right impressions to the buyers who value them highest.
It is important we think about the scale of this challenge. Globally, each SSP manages billions of impressions every day. Those SSPs are integrated with over 100 different DSPs. To keep it simple, let’s consider just 10 Billion impressions. If these requests are sent to 100 DSPs, this equates to 100 Billion bid requests. This is just for one SSP, but this figure is then replicated by 5 to 10 different SSPs, all sending requests to DSPs.
The Need for Smart Throttling
We clearly know that each DSP and each buyer will value impressions differently. They have sophisticated systems and algorithms to decision and attach relative value to each bid request SSPs send them. Given the complexity of each bidding algorithm, inventory optimisation is not as simple as pre-filtering inventory, based on a fixed set of criteria such as geography, publisher and platform. If we filter inventory based on these criteria, we manage the impression explosion, but we ignore the dynamic nature of real-time bidding. We also risk filtering pockets of impressions of value to specific buyers. Campaigns are dynamic, the decisioning criteria evolve and are different for each campaign.
Therefore, we must implement a more sophisticated, automated approach to impression throttling powered by machine learning. At PubMatic we have implemented an automated Smart Throttling product which evaluates every impression based on a range of criteria.
By examining past DSP and advertiser bidding behaviour against the information passed to us in the ad request we forecast the relative value of each request for each DSP. All of this is applied in real time, for every impression.
By examining past DSP bidding behaviour against the information passed to us in the ad request, we forecast the value of each request for each DSP and generate recommendations on which ad requests DSPs would value highest. All of these recommendations are applied in real-time, for every impression.
We know that every DSP will not respond to every bid request. With PubMatic Smart Throttling we send fewer, more targeted bid requests to DSPs instead of sending every impression to every DSP.
our Smart Throttling model is continuously learning, monitoring for changes in buyer behaviour. Meaning it adapts quickly based on buyer bidding patterns to ensure valuable impressions are sent to the buyers valuing them at that point in time.
Machine Learning and Efficient Infrastructure
All this results in PubMatic being able to maintain an efficient infrastructure and reduce the load on our DSP partners. DSPs no longer have to listen to bid requests which represent little value to them, or their advertisers. Prized queries-per-second (QPS) can be devoted to only the highest value impressions which fit their current buying criteria. This is particularly important to DSPs with limited QPS bandwidth.
In a world of DSP and advertiser supply path optimisation (SPO), an SSP who is able to efficiently manage impression volume and QPS is a good partner. Increasingly SPO decisions are being based on this type of criteria, alongside inventory quality, all of which drives revenue for our publishers.
Most importantly, Smart Throttling enables PubMatic to optimise yield for our publishers. By sending impressions to DSPs which are most likely to find them valuable, we can manage greater volume from our publishers. We optimise the QPS we send to DSPs resulting in them delivering higher response rates and typically higher bid eCPMs. Again, driving publisher revenue.
Smart Throttling is just one approach to reducing the volume of low-value impressions from the supply chain. A stubborn focus on inventory quality further improves our platform, ensuring that PubMatic only sends the highest-value inventory to our demand partners. In turn, this drives buyer preference and revenue for our publishers.