When measurement company DoubleVerify (DV) announced plans to acquire AI-focused ad tech startup Scibids for $125 million on Monday, its stock took a dip.
Perhaps investors goggled at the deal price.
But owning AI technology is a logical move for DV as it continues to move beyond measurement into media activation and campaign optimization.
Scibids dynamically adjusts bids for every impression based on an advertiser’s KPIs, such as viewability or desired CPM range. To inform its decisions, Scibids pulls information via APIs from demand-side platforms, including first-party data and media cost, as well as attention data from DV.
“It just takes our data and moves it to an entirely different level of granularity and applicability for our customers,” DV CEO Mark Zagorski told AdExchanger.
For example, DV can use the Scibids tech to create more refined segments for media activation, identify high-attention inventory and optimize campaigns without relying on third-party cookies by accounting for variables such as domain placement, device and geolocation.
Scibids analyzes “millions of possible values,” said Rémi Lemonnier, co-founder of Scibids.
Go with the flow
Although well over half of DV’s revenue comes from programmatic, it previously implemented data for its media-quality segments in a “static” way, according to Zagorski.
Decision-making was also binary: brand safe or not brand safe, fraud or not fraud.
But since Scibids analyzes each client’s data and adjusts its decisions on a per-impression basis, it can take a more fluid approach to prebid campaign optimization.
After running tests with companies ranging from what Zagorski referred to as “mom-and-pop shops” to larger players in the AI algorithm optimization space last year, DV selected Scibids as its partner to jointly launch a tool called the DV Algorithmic Optimizer that combines attention metrics and AI-powered ad decisioning.
That was in June. The tool yielded such strong results that DV decided to buy Scibids less than two months later.
ML at the core
Because DV is essentially a “giant decision engine,” Zagorski said – its primary business is to analyze and classify large volumes of content according to brand safety, suitability and fraud standards – it already has a large data science group and has long invested in machine learning models.
But although the algorithms have advanced and “training is now increasingly being moved to machines,” people are always part of the process, Zagorski said.
Semantic scientists and ontologists evaluate the language that the models train on to ensure it’s not “exclusionary or biased against certain types of content,” he said. These human language experts also use their nuanced understanding of language to guide the models.
Scibids, meanwhile, uses a mix of predictive and generative AI informed by data from customers and “thousands of improvements over the years,” Lemonnier said. The company developed its own models rather than relying on publicly available ones because Lemonnier favors building custom solutions to specific problems rather than modifying already existing technology to fit a different purpose.
The plan is for DV and Scibids to cross-pollinate their teams so DV can learn from Scibids’ expertise. “This is not a rip it out, jam it into our business and drive layoffs” deal, Zagorski said.
Scibids’ roughly 70 employees, including AI engineers and data scientists, will join DV’s team of more than 900 people, bringing the combined company’s headcount to nearly 1,000 total.
