Got Infrastructure Problems? NToggle CEO Explains How Header Bidding Helped His Startup Gain Traction

Header bidding has been great for nToggle.

The startup uses machine learning to shape the bidstream for buying platforms. Instead of looking at every potential impression, nToggle filters out ones it thinks the DSP is unlikely to bid on, reducing the amount of queries by an average of 90%.

Fewer queries to look at translates to better results for buyers and cost savings on infrastructure. The technology came at just the right time. Many DSPs have seen their infrastructure strain to keep up with the flood of impressions coming their way due to header bidding.

“That was the dirty secret of the industry, that the infrastructure was blowing up,” said nToggle CEO Adam Soroca.

NToggle raised a $5.6 million Series A round in February 2015. It’s approaching 20 DSP clients who range from large to small, Soroca said. Clients like DataXu, IgnitionOne, NinthDecimal and Bidtellect pay a monthly software-as-a-service fee to use the platform.

NToggle plans to offer services for the supply side. And it’s looking into “decorating the bidstream” with data, so a buyer can bid on consumers with specific attributes.

Soroca talked with AdExchanger about what he’s up to at nToggle and what he sees as the future of the small DSP market.

AdExchanger: How did you come up with the idea for nToggle?

ADAM SOROCA: Back when I was the chief product officer at Millennial Media, I was trying to convince the company that we had to follow the same playbook as the larger DSPs, which was to expand our data center capacity.

At the time, Millennial had $90 million in cash, and the price tag was $7 million to $9 million just for that wave of expansion. But display DSPs had already spent tens of millions in infrastructure.

I realized I was pushing to write a check for servers, which are a commodity, as opposed to hiring more product managers, engineers and data scientists. I figured there had to be a better way to handle this scale issue for the industry.

How does nToggle solve this infrastructure problem for buyers?

We do that by shaping the bidstream for buyers. Rather than take the firehose, buyers take the most relevant bid requests for their demand. Instead of spending on infrastructure, it allows them to take that money and put that back into innovation.

Outside of maybe three DSPs, the other 447 don’t listen to all the traffic, which is close to 6 million QPS [queries per second]. They can’t afford to. We look at ourselves as democratizing scale and allow any buyer to compete with the largest DSPs.

How does your technology work?

We have a high-speed router with advanced, cutting-edge machine learning. And it’s all controllable through a user interface. Bid requests will travel over our systems and the traffic is evaluated based on all the attributes in the bidstream. We determine what traffic is allowed to go through with machine learning that looks at what will result in a bid or win for the DSP.

What are the top concerns of buyers considering nToggle?

The most common question is, “If you are sitting between buyer and seller, aren’t you introducing latency?” The answer is no. Because our networking connections are dedicated high-speed fiber and we are located next to the supplier end points around the globe, transit time is minimal.

The work we are doing at bid request time is very lightweight since our heavy lifting is done offline. We are not doing the same lookups that a DSP has to do. And our engineers come from fin tech and ad tech backgrounds and use pretty clever tricks to make sure the systems run efficiently. Our system adds two milliseconds.

What’s header bidding done to business?

That has been fantastic for us, as you might imagine, because of the incredible growth in traffic out there. QPS management was a burden even before header bidding.

As header bidding has exploded it’s brought companies’ infrastructure to their knees. We shape traffic based on whether or not an ad ends up serving on a page, so we’re able to throttle down volume from much of the header bidding traffic that don’t result in revenue.

How do buyers evaluate whether nToggle is working?

They will look at revenue per thousand QPS and overall revenue growth. That shows the concentration or richness of the bidstream we provide. And they look at how much money have we saved them on infrastructure. We are 20% to 30% of the cost of listening to traffic on their own.

Buyers also want us to evaluate more traffic than they were before, because buyers want growth. They will see two times, five times or ten times more traffic than before, and shrink it down to a lower QPS.

What’s ahead in the next year for nToggle?

We’ve been pulled into the supply side. Header bidding is crushing their systems as well. We have pilots with supply partners to manage their outbound QPS.

We evaluate the outbound requests before it leaves the SSPs’ data centers to ensure buyers see more relevant supply. And they are looking at the cost savings as well, because every time a bid request leaves a data center it costs money.

And, we return a “no bid” faster, in half a millisecond, if we already know it will not turn into money, versus 100 to 300 milliseconds. That allows exchange servers to speed up and be more efficient. And we are investing in data enrichment.

What can nToggle do with data enrichment?

Just like supply access is a commodity, so is access to third-party data. A buyer has to go do a custom integration, and there are dozens of connections out there. Even the big guys can’t keep up with the changes companies make when a new signal is added, and small to medium-sized buyers can’t do it all.

We are currently in a beta with a couple of companies where we are toggling against the data, fraud and viewability signals, where we can filter the traffic before it hits a buyer’s data center, versus a pre-bid integration.

If nToggle helps solve the infrastructure problem for DSPs, do small ones still go out of business?

A generic programmatic bidder is not going to win. But many of these DSPs will find homes. Publishers are going to buy DSPs, like Time Inc. buying the platforms Viant and Adelphic. Advertisers like Netflix and LinkedIn have their own DSP, and you have gaming companies in the mobile space that have built or bought DSPs.

But the notion of a small, standalone DSP? That won’t prevail.

This interview has been condensed and edited.


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>