Somewhere around midafternoon on a Tuesday, a media buyer at a mid-sized agency notices something off. A campaign on The Trade Desk is pacing ahead of schedule, but the numbers in Campaign Manager 360 don't quite match. Meta's dashboard shows a different cost-per-acquisition entirely. What follows is familiar to anyone who's worked in programmatic: a multi-tab excavation across platforms, exporting CSVs, cross-referencing taxonomy that refuses to align, hunting for the discrepancy's source.
It's the kind of operational friction that platforms haven't solved, mostly because the problem lives in the spaces between them.
Kovva, a startup that emerged from closed beta on May 18 with a team of exactly four people, is betting that AI agents can work in those gaps. The pitch is direct enough: autonomous agents that live where media buyers already spend their day—Slack threads, email inboxes, web dashboards, Teams and Zoom calls—and handle the cross-platform coordination that no single vendor has bothered to automate.
"AI has been built into the platforms," the company says, "but not between them."
When Platforms Don't Talk
In practice, what Kovva is offering looks something like this: real-time Q&A threaded directly into Slack conversations (a buyer asks about pacing mid-discussion, the agent pulls live data from three DSPs and answers inline), discrepancy analysis that doesn't just flag problems but suggests fixes, budget reallocation recommendations that factor in return-on-ad-spend across Meta, YouTube, and DV360 simultaneously. The agents can even draft client updates when a trader is out of office—though everything requires human approval before it ships.
At launch, Kovva claims north of 50 integrations: demand-side platforms like The Trade Desk and DV360, social networks including Meta and TikTok, ad servers, measurement vendors, productivity tools. The agents pull data from these systems, normalize the taxonomy—because "campaigns," "ad sets," and "insertion orders" all mean subtly different things depending on where you're standing—and surface insights or flag risks before they metastasize into bigger problems. Every recommendation, the company says, comes with an explanation and a citation.
The use cases skew toward the operational weeds: campaign setup and quality assurance (validating pixels, checking for missing click macros), creative fatigue detection, pacing alerts, the kind of configuration-level optimization that requires stitching together data from four platforms at once. Co-founder and COO Tanja Mimica framed the gap plainly in an interview with AdExchanger published Monday: "Inside platforms it's AI everywhere—but in between, buyers are still in spreadsheets."
There's something almost understated about that observation, though perhaps it shouldn't be. The ad tech industry has poured resources into making individual platforms smarter—better algorithms, faster bidding, more granular targeting. But the connective tissue? That's been left to junior traders armed with VLOOKUP formulas and institutional knowledge about which metrics map to which columns in which export.
The Team Behind It
All four founders share a history. Kyle Dozeman, now CEO, spent time most recently as chief revenue officer at PubMatic. Mimica, along with co-founders Hassett and Mueller, previously built Martin, a DSP that PubMatic acquired in September 2022. After the deal closed, Mimica became VP of PubMatic Activate, Hassett took over engineering leadership, and Mueller worked as lead data scientist. Hassett's background includes earlier stints as an AI engineer at Tesla and Rio Tinto—the kind of résumé that suggests someone comfortable with complex systems and large-scale automation.
The company is self-funded, at least for now. Dozeman told AdExchanger they're planning a fundraise later this year, targeting late summer. Kovva lists a Cincinnati address in its site footer; legal pages were updated as recently as late March.
A Crowded, Suddenly Urgent Market
Kovva's launch arrives in the middle of what can only be described as an agentic AI land rush across ad tech. The Trade Desk introduced Koa Agents in late April—backed by Stagwell—for planning, buying, optimization, and measurement. Magnite followed with its own buyer and seller agents days later. Omnicom has been testing agent-to-agent buying, aiming to compress the supply chain. NBCU and agency RPA reportedly completed what they characterized as the first live agentic-to-agentic linear TV deal earlier this year.
The timing isn't coincidental. Standards are starting to coalesce: IAB Tech Lab's AAMP protocol for agentic ad buying entered open beta in April, and Model Context Protocol-style interfaces are proliferating fast. Comcast connected its Universal Ads platform to AI via MCP in mid-May; Channel99 did something similar for B2B marketing data around the same period.
Dozeman, for his part, described Kovva to AdExchanger as "connective tissue," not a platform replacement. The agents are designed to sit on top of existing infrastructure and handle the work that falls through the cracks—the discrepancies, the cross-channel questions, the QA checks that don't fit neatly into any single platform's workflow.
It's a reasonable positioning, if the market plays out that way. But it also raises a question: if The Trade Desk, Magnite, and the holding companies are all building or deploying their own agents, what prevents them from eventually filling the gaps themselves?
What Remains Unclear
Pricing isn't public yet. Neither are customer names or detailed case studies, though AdExchanger referenced feedback from early testers. The degree to which Kovva's agents can execute changes programmatically—versus simply drafting recommendations for human review—isn't entirely spelled out. Setup assistance and QA are described, but the boundaries of write-access permissions remain vague.
Security and compliance details are similarly thin on the public site: a general privacy policy, but no mention of SOC 2 certification or data residency specifics. For a product that will presumably touch sensitive campaign data and client information, those details matter.
For a four-person team entering a market where much larger players are moving aggressively, the challenge will be demonstrating that the in-between layer has durable value—and maintaining 50-plus integrations as platforms evolve their own AI capabilities. If media buying remains as fragmented as it is today, there may be room for both approaches. If not, well. The history of ad tech is littered with middleware that seemed essential until it suddenly wasn't.
