It’s the great irony of programmatic.
The bidding is automated down to the millisecond, but the day-to-day work of running campaigns is still largely manual, from copying data between systems and reconciling reports to running QA checks and drafting explanations for clients when metrics don’t line up.
“Inside of the platforms, it’s AI everywhere – but in between, buyers are still in spreadsheets for most of the day,” said Tanja Mimica, COO and co-founder of Kovva, a new AI ad tech startup that came out of closed beta on Monday.
New AI ad tech startups are coming out of the woodwork, but Kovva is staking out a less glamorous niche than most: using agents to deal with the cross-platform chores that have mostly resisted automation.
Because, hey, the less sexy the work, the more valuable it often is.
“The idea is that a media buyer can delegate their workflows,” said James Hassett, Kovva’s CTO and a co-founder, “and we’re concentrating on the ones that cause operational pain, the really tedious ones.”
Bye bye, busywork
Kovva’s four co-founders all worked together in and around PubMatic.
Mimica and Hassett co-founded Martin, the DSP that PubMatic acquired in 2022, which became the basis for Activate, its direct-to-buyer solution cut from the same cloth as Magnite’s ClearLine and The Trade Desk’s OpenPath. After the acquisition, Mimica served as VP of Activate and Hassett became PubMatic’s VP of engineering.
Andrew Mueller, meanwhile, Kovva’s VP of data science, was Martin’s lead data scientist, and Kovva’s CEO is Kyle Dozeman, PubMatic’s freshly former CRO.
The idea for Kovva grew out of their mix of experience building a DSP, working inside PubMatic on the sell side and running trading teams. It got a little frustrating watching buyers burn out on grunt work, Mimica said.
“All of this operational drag really piles up,” she said. “We’re trying to shave off as much of this busy work as possible, so buyers can focus on the strategy and the clients.”
Don’t hate, delegate
Kovva’s AI agents piggyback on a trader’s existing tools and are designed to function less as chatbots and more like team members who don’t mind doing scutwork.
Buyers can interact with them in Slack, via email or through a web interface. The point is to hand off work, not sit there typing prompts.
One early focus is quality assurance. A few hours after a campaign goes live, for example, a Kovva agent can run through the sort of checklists that traders usually have to do by hand to make sure everything’s kosher. Is the campaign running on the right geos? Are the pixels placed correctly? Is it pacing properly?
Tracking down cross-platform discrepancies is another use case. If The Trade Desk is reporting one thing and Google is reporting something else, investigating that inconsistency can be enormously time-consuming.
“You’ve got to look across attribution windows and time zones and there are reporting delays – all of these annoying things,” Hassett said.
Instead of a trader spending hours picking through settings and exports, Kovva can do the first pass, come back with an explanation for the issue and, where possible, a suggested fix.
The agents can also pitch in with recommendations on budget allocation. If, say, a campaign looks likely to underdeliver or a client has extra money to spend, Kovva can propose where to move budget across platforms based on attribution data from MMM and MTA tools.
And then there are more advanced features, like creative fatigue detection. Rather than ranking ad creative based on outcome, Kovva models the saturation effect and can flag when it’s time to swap out a particular creative or maybe show it to a different audience.
The list goes on.
Some of the workflows weren’t even on the road map at first, Hassett said, and came from suggestions by early testers, like agent-based client support for when a trader is out of the office. In a case like that, Kovva can pull in information and reports from across platforms and automatically draft answers to a client’s email, with a human required to approve it before it’s sent.
The translation layer
Everything Kovva’s agents do depends on having a view across systems. You can’t reallocate budget from one platform to another or compare performance without normalizing the data flows.
It’s got around 50 integrations so far, spanning DSPs, ad servers, social platforms and measurement products, with plans to add a lot more, Hassett said.
“One of the value propositions here is that we also maintain all of the integrations,” he said, “which is part of what stops large agencies from trying to build some of this themselves.”
But being plugged in is only useful if agents can make sense of what they’re pulling in. They need a standard taxonomy. Yet different platforms describe similar objects – campaigns, ad sets, metrics line items, insertion orders, etcetera – in different ways.
Which is why so much of Kovva’s engineering work has been on creating a system for mapping naming conventions, lining up structures and creating a common language its agents can use across platforms.
That work also shapes how Kovva thinks about its place in the ecosystem, which is as connective tissue, not a replacement for what buyers already use.
“There’s not a clear set of competitors or pieces of tech we’re looking to replace or compete with,” Dozeman said. “We’re just trying to be additive and incremental.”
For now, Kovva’s entire team is just the four co-founders, and the company is self-funded. They plan to bring on more engineers over the next few months and start raising capital in late summer.
