By Making The Internet Faster, Cox Boosts Programmatic Revenue

dyn-and-coxCox Interactive is piloting smart routing technology from Dyn that speeds up the outbound bid request to demand-side platforms (DSPs). It unveiled early results of the test at AdExchanger’s Programmatic I/O Ops Talks in New York on Wednesday.

Faster connections mean more DSPs can deliver bids before publisher-set timeouts. And more bids mean more programmatic revenue for both Cox and the 5,000 publishers in its ad network.

The smart routing technology uses Dyn’s vast knowledge of network connections. Last Friday’s DDoS attack against Dyn served as a testament to its scale.

As a domain name server (DNS) provider, Dyn claims to traverse every internet pathway twice a minute. Like the traffic app Waze, Dyn can observe where slowdowns occur and use tech to reroute a request to a faster pathway.

In the early tests, results varied dramatically by DSP. Dyn helped one DSP run 20 milliseconds faster, for example, which improved revenue by only 3%. Another DSP ran 15 milliseconds faster, boosting revenue by 150%.

“Most DSPs have a huge potential for improvement, if you can drag response times to the critical threshold,” said Scott Siegler, GM and VP of engineering for Cox, which tested the tech with more than 30 DSPs.

The average DSP might see a large portion of bids come back just after Cox’s timeout period, so just a little push can bring in requests for most of their impressions below the timeout. Using the tech to connect to extremely slow or fast DSPs doesn’t improve revenue as much since those DSPs were already missing or making the timeout by a large buffer.

The test with Cox is only at a proof-of-concept stage, said Scott Brazina, Dyn’s VP of media. The former DataXu CMO joined Dyn to help the DNS provider use its extensive data of network connections for a new market: media and mar tech.

Programmatic auctions could use the extra help with speed now more than ever. Header bidding offers more inventory programmatically, flooding DSPs with bid requests and slowing them down. Overall, programmatic is on the rise, forcing DSPs to up their capacity even further.

With the tech, Cox’s algorithms have enough time to pick where to route an impression, lowering the burden on DSPs and speeding up their response side, Siegler said.

With the extra speed, Cox will allocate saved milliseconds to optimize its own operations better. Siegler’s team has created “cascading algorithms” that decide how to price impressions and where to route them.

“The more complex algorithms perform better, but take more time to run,” Siegler said.

With faster connections, Cox can run the more complex algorithms more frequently. Cox also vets impressions with inventory quality providers, such as Forensiq or Pixalate. Having more time means Cox can do more extensive filtering.

The approach requires only one side to buy in, avoiding extensive conversations about collaboration. But the tech does have limitations.

It can send a bid request faster, but it can’t speed up the return path from the DSP (though Dyn is working on some ideas for how to influence that). And users like Cox must run their own servers for the technology. Ad tech or media companies that use the cloud will be out of luck for the moment, though Dyn plans to start conversations with Amazon Web Services and other cloud providers.

Once it finishes the proof of concept with Cox, Dyn plans to seek other customers in the DSP and supply-side platform space. Dyn also wants a test customer in the APAC or EMEA regions. Since network signals take hundreds of milliseconds to travel across the world. Brazina says customers in global markets will see the biggest benefit.

But it’s still early days for Dyn, which hasn’t yet worked out a pricing model for the tech. Brazina offers one promise: “Both sides are trying to make it about the ROI of the product.”

 


Popular On AdExchanger Right Now:

 

Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>