Many programmatic platforms are in a bind. Internally, they’re under pressure to reduce infrastructure costs. Externally, advertiser clients are demanding that they lower their fees. George Levin, co-founder and CEO of demand-side platform [DSP] Getintent, spoke with eMarketer’s Ross Benes about how programmatic firms have to adapt to the ad industry’s changing imperatives.
Header bidding changed how programmatic auctions work because it allowed many supply-side platforms to simultaneously bid on the same piece of inventory. As header bidding has become more popular, have you seen an uptick in the amount of bids or queries you have to process?
Definitely. It’s growing very fast because of header bidding. At peak times, we are at 2 million queries per second. In 2015, it was something like 300,000. It’s hard to draw a direct cause and effect because we have more clients now. But overall, I think maybe 50% of the increase [in bids per second] is because of header bidding. But that’s a rough estimation because it’s very hard to set accurate parameters.
If you are processing more bids per second, theoretically that means you are burning through more central processing units [CPU], which could wear out your servers and possibly drive up electric bills. Has the uptick in bids per second been costly?
For us, maintenance cost is a very small share of our expenses. We build out infrastructure in a very efficient way. I assume that for the major DSPs, [the increase in bids per second] could be a very big problem.
What’s different about your infrastructure setup compared with other DSPs?
We use quite cheap hardware servers instead of the cloud. Basically, it requires more people to maintain because it’s more sophisticated than just keeping everything stored with Amazon or Google, but it’s still 10 times cheaper.
Programmatic has become so complicated to the point that DSPs frequently unintentionally compete against themselves when they bid on inventory. How do you avoid this?
We have our own algorithm to avoid [duplicate bids], but it’s just the beginning of this. Basically it’s win-rate optimization. The idea is that we can use historical data from advertisers and based on this data, our artificial intelligence creates this probability model. For each impression, we know the probability of who is most likely to buy this impression.
A year ago, 20% [commission] was very OK. But right now, it’s very hard to sometimes get 10%. People want to pay less.
Ever since The Guardian sued Rubicon Project for its undisclosed fees, vendors have been under real pressure to lower their fees. Have you had to change your fee structure?
Yes, it’s going down every quarter because competition is tough. Commission-based pricing [rates in the ad industry] are going down.
How much have your commission rates changed?
A year ago, 20% was very OK. But right now, it’s very hard to sometimes get 10%. People want to pay less.
If clients demand that your commission rates run no higher than 10%, how do you keep growing your business while cutting fees?
Our average check is growing and we’re getting bigger clients. If we onboard someone big, even if they only pay us 8%, the absolute numbers still give us big revenue. So the goal is to grow the average check and to onboard more big clients.
Last year, the Interactive Advertising Bureau Tech Lab released ads.txt as a tool for the industry to fight domain spoofing. Many DSPs have built ads.txt dashboards into their products for advertisers to use. Did you have to do this?
Yes. It’s on the campaign-management level, so agencies can easily use it when they set up new campaigns. It’s a very popular feature; all the agencies use it.