The advertising technology world has a gaps problem—not in what platforms can do, but in how they talk to each other. Or rather, how they don't.
That's the premise behind Kovva, at least. The startup, which emerged from closed beta the week of May 18, is the work of four founders who spent years inside PubMatic's programmatic infrastructure before deciding the real opportunity wasn't optimizing within platforms, but bridging between them. Their product: AI agents designed to handle the operational mess that lives in those gaps—QA checks, discrepancy hunts, cross-platform budget shuffles, creative fatigue warnings.
Kyle Dozeman, Kovva's CEO, left his role as PubMatic's chief revenue officer for the Americas in April to lead the company. Tanja Mimica, now COO, founded Martin—a DSP and measurement platform PubMatic acquired in September 2022—and later ran PubMatic Activate as a vice president. James Hassett, the CTO, co-founded Martin with her and held a VP engineering role at PubMatic. Andrew Mueller, the data science lead, brings 15 years in AI and machine learning across adtech, including time as Martin's lead data scientist.
For now it's just the four of them, self-funded, working from a suite at Rookwood Exchange in Cincinnati. Engineers will come in the next few months. A fundraise is planned for late summer 2026, according to an AdExchanger interview published May 18.
Whether buyers actually want this kind of connective tissue—or whether it just becomes another layer to manage—is the question the team will spend the next year answering.
What the Agents Do (and Don't)
Kovva positions itself not as a replacement for existing platforms but as the layer between them. The agents operate across more than 50 integrations: The Trade Desk, Display & Video 360, Campaign Manager 360, Google Ads, Meta, Amazon, Reddit, TikTok, Walmart Connect. Plus the usual office suspects—Slack, Zoom, Teams, Outlook, Excel, PowerPoint.
The use cases cluster around operational work that's tedious enough to resist automation but not important enough to justify full-time headcount. Post-launch QA happens within hours: geos, pixels, pacing, macros, tracking. Discrepancy investigations that span DSPs, ad servers, and analytics platforms—attribution windows, time zones, reporting delays, the kinds of problems that eat up Tuesday afternoons. Budget allocation recommendations across Meta, YouTube, The Trade Desk, DV360. Creative fatigue detection that runs quietly in the background.
Early testers have used the agents to draft client support emails. Human approval required before sending, naturally.
The interface meets buyers where they already live: Slack, email, Zoom, Teams, or a web app. Ask about campaign performance in Slack and the agent pulls historical context, flags anomalies, proposes root causes. Need a quarterly business review? The agent generates narratives, charts, tables, executive summaries. Translating a media plan into platform configurations, uploading creatives, validating settings—workflows that can, in theory, be delegated.
Every output includes an explanation and citation, according to the company. The emphasis on transparency reflects what the team calls "augment, don't replace." Buyers stay in control; agents handle the tedious parts.
Which sounds reasonable enough, until you consider how many times the industry has heard similar promises.
The Taxonomy Problem
What makes this more complicated than a chat interface is the normalization layer. Kovva maintains integrations and maps taxonomy and naming conventions across platforms—infrastructure necessary for agents to act intelligently across systems that don't share data structures.
That's not glamorous work. It's also not optional if you want agents to do anything useful across, say, The Trade Desk and Meta without constant human translation.
The team emphasizes "buyer-first" adoption: the product is built for media teams at agencies and brands, not platform vendors looking to lock in workflows. Whether that positioning holds as the company scales and potentially takes outside capital is another matter.
A Crowded Moment for Agentic AI
Kovva is entering a market that has moved fast toward agentic workflows over the past year or so. The Trade Desk launched Koa Agents for media planning, buying, optimization, and measurement. Yahoo DSP added agentic capabilities with bring-your-own-model and API interoperability in January. Magnite expanded agentic tools for buyers and sellers in April. Kargo announced Project KERA, an agentic media-buying and creative engine, in late March, entering closed beta with select partners. Kochava opened its StationOne workspace and began testing the IAB Tech Lab's AAMP protocol, building infrastructure for cross-platform agent coordination. Invoca launched AI agents tied to paid media pipelines around the same time.
The pattern reflects a shift in how platforms think about automation. Earlier AI features often focused on optimization within a single walled garden—better bidding, smarter targeting, more effective creative selection. Agentic workflows assume coordination across systems, with agents acting as intermediaries that understand context from multiple platforms and execute tasks that span them.
Whether buyers will trust that coordination remains an open question. The industry has cycled through enough AI hype cycles to know that adoption depends on transparency, reliability, and whether the tools actually reduce operational burden—or just add complexity with a new interface.
The Pivot from Sell-Side to Buy-Side
The founders' shared history at PubMatic gives Kovva technical grounding in programmatic infrastructure. It also raises positioning questions. PubMatic is a sell-side platform; Kovva targets buy-side workflows. The Martin acquisition in 2022 brought DSP capabilities and measurement tools into PubMatic's portfolio, and Mimica's work leading PubMatic Activate—built on Martin's infrastructure—gave the team experience bridging buyer and seller workflows.
Even so, the move from a publicly traded adtech company to a bootstrapped startup represents a specific bet: that the gap between platforms is large enough, and persistent enough, to support a dedicated product. That media buyers spend too much time reconciling data, troubleshooting discrepancies, and translating decisions across systems. And that AI agents can absorb that work without replacing the judgment calls that still matter.
It's a reasonable thesis. Proving it will require more than clever integrations.
What Comes Next
No pricing or trial details are public. The website invites visitors to "get started," but early access appears to be managed through direct outreach. The plan to raise capital later this summer suggests the team is focused on product development and early customer validation before attempting to scale.
In an industry where platforms have spent years trying to keep buyers inside their own ecosystems, Kovva's pitch is that the real value lives in the connective tissue. The operational glue. The unglamorous middle layer where campaigns actually get built and maintained.
Maybe that's where the next phase of adtech automation happens—not within platforms, but between them. Or maybe it's just another coordination problem waiting to be solved by the next wave of tools.
The four founders in Cincinnati are betting on the former.
