Tanja Mimica of Kovva on What the Future Holds for Media Buyers

Kovva's Pitch: Real Pain, Modest Business

Tanja Mimica, founder of Kovva (an AI workflow tool for media buyers), opens with numbers that will feel familiar to anyone inside a holdco: 72% of agencies cut prices in 2025, 70% of revenue goes to headcount, net margins at 13% and falling. The direction is real even if the precision is arguable, and that pain is the problem Kovva is selling into.

The positioning is deliberate. Kovva lives in trafficking, pacing, QA, and discrepancy checks. It explicitly stays off strategy and planning. That keeps it out of direct competition with The Trade Desk (the largest independent ad-buying platform), Google's DV360, and Meta, all of which are building AI inside their own walls. Mimica picked the space between the platforms because the platforms themselves are owned. Smart. But you can't charge much for glue, and that's what this is.

The demo is genuinely useful. A client emails asking for a cross-platform report; Kovva reads the request, pulls the data, and drafts a response in an hour instead of days. The thing that breaks first is the connectors. Every DSP, ad server, and verification tool changes its API and reporting schema on its own schedule, and discrepancy checks are only as good as the mappings underneath. At 90 days you're not fighting the AI, you're fighting FreeWheel's export format and a search platform that renamed three fields.

The bigger conceptual bet is what Mimica calls the "compounding agency": knowledge captured automatically from the work itself, applied everywhere, widening the gap month over month. The graveyard she cites includes intranets, wikis, Confluence, and Salesforce, all dead because nobody owned keeping them current. Her answer is passive capture. That is also exactly the promise every knowledge-management vendor made for 20 years. What has to be true for this to work is that agencies have durable, encodable operating philosophy that survives client churn and staff turnover. In a business where accounts move and traders leave, that durability is the assumption nobody has proven.

There are two different customers here, and they want different things. The agency P&L owner wants margin relief and will automate pivot tables happily. The junior media buyer whose day IS pivot tables hears "AI frees you up for strategy" and knows better. Mimica says the quiet part: ad-ops people freed from spreadsheets don't automatically become strategists, because strategy seats already have people in them. Agencies buying this need to plan the redistribution deliberately, or they'll gut their own talent pipeline and discover in three years they have no seniors because they hired no juniors.

The CFO math is straightforward. If 70% of revenue is headcount and ops is a slice of that, a tool that makes each buyer 20% faster but keeps everyone employed is a cost. The payback shows up only if Kovva lets an agency win a pitch on price it would otherwise lose, or absorb a new client without adding headcount. That's measurable. "Compounding intelligence" is not. Buy the workflow time savings; ignore the philosophy premium until it shows up as a won pitch or an absorbed account.

Publishers and SSPs get nothing here. This is a buy-side story about who owns the automation layer.

Our call: By the 2027 agency budget cycle, at least one of The Trade Desk, DV360, or Meta ships or acquires a cross-platform workflow feature that reaches outside its own inventory, squeezing standalone orchestration players like Kovva. The whitespace Mimica is selling into is exactly what an incumbent closes once it sees demand, because a buyer orchestrating across platforms from The Trade Desk's console has less reason to leave it. A small product extension for a platform is a large lock-in payoff. Kovva has real demand and genuine whitespace, but the position between walled platforms limits pricing power, and the compounding-knowledge claim is unproven on top of a real workflow tool.

The Market Analyst. Mimica cites 72% of agencies cut prices in 2025, 70% of revenue still goes to headcount, and net margins sit at 13% and falling. Take the precise numbers with salt, she's selling to the pain she's describing, but the direction is real and everyone in holdco land knows it. The interesting part is where she is NOT competing. She explicitly stays off strategy and planning and lives in trafficking, pacing, QA, discrepancy checks. That is a deliberate dodge of The Trade Desk, DV360, and Meta, all of whom are building AI inside their own walls. In plain terms: she picked the hallway between the rooms because the rooms are owned. Smart positioning, weak pricing leverage. You can't charge much for glue.

The Skeptic. The "compounding agency" thesis is the load the whole pitch carries, and it's the shakiest part. Mimica argues an agency's intelligence can compound because knowledge gets captured automatically and applied everywhere, widening the gap every month. Maybe. But she also lists the graveyard of every prior attempt: intranets, wikis, Confluence, Salesforce, all dead because nobody owns keeping them current. Her fix is that the AI captures knowledge passively from the work itself. That is exactly the promise every knowledge-management vendor made for 20 years. What has to be true for this to work: agencies have durable, encodable "philosophy" that survives client churn and staff turnover. In a business where accounts move and traders leave, that durability is the assumption nobody has proven.

The Operator. Tuesday morning, the client emails asking for a report. Mimica's demo has Kovva reading the request, pulling cross-platform data, and drafting a response in an hour instead of days. Good. The thing that breaks first is the connectors. Every DSP, ad server, and verification tool changes its API and its reporting schema on its own schedule, and discrepancy checks are only as good as the mappings underneath. At 90 days you're not fighting the AI, you're fighting FreeWheel's export format and a search platform that renamed three fields. The second thing that breaks: someone has to write the Playbooks. Mimica admits knowledge management dies because that job belongs to no one. Encoding the agency's best buyer worldview is that same unfilled job, and it still belongs to no one.

The Customer / End User. Two customers here, and they don't want the same thing. The agency P&L owner wants margin relief and will happily automate pivot tables. The junior media buyer whose day IS pivot tables hears "AI frees you up for strategy" and knows better. Mimica, to her credit, says the quiet part: ad-ops people freed from spreadsheets don't automatically become strategists, because strategy seats already have people in them. For the buyer, this product means fewer entry-level ops jobs, not a promotion. Agencies buying this need to plan the redistribution deliberately, or they'll gut their own talent pipeline and discover in three years they have no seniors because they hired no juniors.

The CFO. The pitch lands squarely on 13% margins, so run the actual math. If 70% of revenue is headcount and ops is a slice of that, orchestration only pays back if it removes heads or lets you take on accounts without adding them. A tool that makes each buyer 20% faster but keeps everyone employed is a cost, not a saving. The real payback question is whether this lets an agency win a pitch on price it would otherwise lose, or absorb a new client with the same team. That's measurable. What's not measurable, and what CFOs should refuse to pay for, is "compounding intelligence" with no line on the P&L. Buy the workflow time savings. Don't pay a premium for the philosophy.

Where the council splits:

The Market Analyst likes the "between the platforms" position; the CFO sees that same position as the reason Kovva can't charge enough to matter. Owning the glue is defensible and cheap to attack, which is exactly why glue is a hard business.

The Skeptic and the Operator agree the compounding thesis rests on someone maintaining the knowledge, and both doubt that someone exists. Mimica's answer, that the AI captures it passively, is the one claim that would make or break the whole category, and it's the one thing the episode doesn't prove.

What this hinges on: whether cross-platform orchestration produces margin relief you can see in the P&L, or just adds a vendor line. The council leans toward "real demand, modest business." The pain is genuine, the whitespace is genuine, but the position between the walled DSPs limits pricing power, and the compounding-knowledge claim is unproven marketing on top of a real workflow tool. Before committing, an agency should pilot one workflow, measure hours actually removed against the subscription, and ignore the compounding story entirely until it shows up as either a won pitch or an absorbed account with no new headcount.

Impact on the broader ecosystem is indirect. Publishers and SSPs get nothing here. This is a buy-side story about who owns the automation layer, and the answer is still up for grabs.

Prediction: By the 2027 agency budget cycle, at least one of the major buy-side platforms building AI inside its own walls (The Trade Desk, Google's DV360, or Meta) will ship or acquire a cross-platform workflow feature that reaches outside its own inventory, squeezing the standalone orchestration startups.

Confidence: Medium — the whitespace is obvious and the incumbents have every reason to close it.

Confidence: Medium — incumbents see the same whitespace Mimica is selling into, and they have every reason to close it.

Why: Mimica's whole positioning rests on the DSPs building AI only within their walls, leaving the space between platforms open. That gap is precisely what an incumbent closes once it sees demand, because a buyer who orchestrates across platforms from The Trade Desk's console has less reason to leave it. The Trade Desk already rebranded its AI push and Google and Meta are pouring money into buy-side automation, so extending one step outside their own reporting is a small product move with a large lock-in payoff. The opposite outcome, incumbents staying politely inside their walls while startups own the connective layer, is the less likely one because glue between rivals is worth more to the platform that captures the buyer than to a neutral third party.

Revisit by 2027-06-30: We're right if The Trade Desk, DV360, or Meta ships or buys a workflow tool that pulls and acts on data from a competing platform. We're wrong if all three keep their AI automation confined to their own inventory and reporting through mid-2027.