Rocket Fuel intends to release a free anti-fraud product in June called Bot Or Not – available to advertisers regardless of whether they’re clients. (Update 5/27: The company has settled on the name Botfinder.)
CTO Mark Torrance announced the product’s existence at the Rocket Fuel Summit in New York City on Tuesday and shined a little light into Rocket Fuel’s black box to provide some high-level insight into how its technology works.
Evil Bots
Fraud is increasingly in the limelight and tech companies are doing more than talking. Consider Google’s purchase of spider.io and Integral Ad Science’s purchase of Simplytics.
While Rocket Fuel works with fraud-detection partners, Bot Or Not is being developed in-house. “It almost works now,” Torrance said. “This is a very hard problem to solve. We need to ensure the quality of inventory we deliver to a marketer’s campaign is as clean as possible.”
Bot Or Not is a bit of JavaScript code that can be placed on a website, designed to determine the amount of fraud that occurs on a given pixel, identifying whether it’s – to be literal – a bot or not.
There’s still some tinkering that needs to be done, Torrance said. “If you have very little evidence of a user – maybe we haven’t seen them very much across the Web or they have a new browser – they might be ambiguous and (Bot Or Not will) score them as a bot. We’re still working on how a customer should set the threshold to the right range.”
Opening Skynet
Torrance also explained how Rocket Fuel’s technology works.
“The approach we take at Rocket Fuel is called artificial intelligence models,” he said. “The problem we’re trying to solve is what chance this customer will convert if I show him an ad right now.”
AdExchanger Daily
Get our editors’ roundup delivered to your inbox every weekday.
Daily Roundup
He compared this with the way humans tend to segment their media-buying decisions. An ad designed for retargeting might go for X amount of dollars; one for a CPG brand might go for Y amount.
Rocket Fuel, Torrance said, takes a bunch of different factors – whether the website is good, whether the impression will be delivered at a good time of day, past purchase behavior, etc. – and tallies it up to determine the likelihood, in terms of percentage, that the consumer will convert.
While human ad buyers might negotiate the price in advance, typing the price into a box and tweaking bids to get the correct delivery, Rocket Fuel’s algorithms are designed to buy based on likelihood of conversion and tries to determine the optimal mix of high-priced impressions and lower-priced ones. Through this, Rocket Fuel hopes to achieve the best possible ad campaign performance.
“We might bid more for a few choice impressions we think we’ll convert,” Torrance said. “We trade that off with impressions we buy at lower prices. The execution isn’t up to us or any people; it’s up to the software agents running in the computers.”
Rocket Fuel’s technology is also designed to examine every impression it purchases, whether pricey or not. It predicts how well each impression will lead to a conversion and compares that prediction with the impression’s actual performance.
“When we see the conversion rate is out of bounds, that’s a problem, a bug we have to work on,” Torrance said. “We’re not guessing which impressions to buy, we’re evaluating every opportunity. All of these opportunities get scored and from that score we determine a price.” Higher scores – that is, impressions with a higher likelihood to generate a conversion – the higher Rocket Fuel will bid.
Torrance also discussed the ad tech companies that focus strictly on retargeting. (Criteo, which like Rocket Fuel had a high-profile IPO, comes immediately to mind, though Torrance didn’t call it out by name.)
“Many think retargeting is an easy or a different kind of problem, so there are companies that specialize in it,” Torrance said. “But at Rocket Fuel, we think it’s a similar kind of problem.”
It makes sense for Torrance to address retargeters, since their presence siphons client ad budget away from Rocket Fuel. His pitch: “When companies are not using these AI models to decide where their spend should be for retargeting, what they’re doing is paying the same price, bidding the same amount for this different type of traffic.”
Torrance claimed it’s best to view retargeting and non-retargeting advertising as similar activities, “rather than splitting the budget on an ad hoc basis.”
