Kawaja: AI uncertainty freezing ad-tech M&A; decisioning plus unique signal will win

Full analysis

Terry Kawaja of LUMA Partners, the banker who runs most of ad-tech's M&A, stood up at the Prebid Summit and told the room why nobody is buying anything: nobody can tell whether a company's AI is a real asset or a feature that Anthropic erases with its next model. His bet on what survives the freeze is "decisioning coupled with unique signal." In plain terms: AI that makes the buying decision, sitting on ad-targeting data nobody else has.

This is easy to undo as a read on the market, hard to undo as a strategy. A corp dev team can wait out a frozen deal market. A company that spent two years building a product on top of someone else's model cannot un-spend that. What's actually being decided is where the few checks that do get written in 2027 will go. Nothing sets a hard deadline, but every frontier model release resets the clock on how buyers judge "defensible."

The Market Analyst. Kawaja is describing a repricing, not a pause. When a buyer can't tell whether your moat survives the next model drop, they don't pay a premium for it, they underwrite it to zero and make you prove it later. That is why workflow-automation ad-tech, the copy generators and generic bid optimizers, will see multiples compress regardless of revenue. The money that moves flows to companies sitting on data a model can't synthesize: licensed identity, retail-media purchase data, clean-room assets. For a non-specialist: investors stopped paying for clever software and started paying for data nobody can copy. Expect distressed acqui-hires of the signal-thin players before any marquee deal prints.

The Skeptic. Kawaja earns his living when deals close, so "AI uncertainty froze the market" is a convenient story that lets everyone off the hook. Rates, credit, and three years of ugly public ad-tech prints froze this market long before Anthropic shipped anything. And "decisioning plus unique signal" is not a 2027 insight. The DMP vendors said it in 2015. Megan Clarken's Criteo said it. It is the oldest moat in the business, dressed in new clothes. What has to be true for this to be the winning frame: signal scarcity persists, cookies stay dead, and no walled garden solves targeting natively for buyers. All three are shaky. For a non-specialist: the banker says the freeze is about AI because "the economy is bad" doesn't book him any fees.

The Operator. The freeze is already on your desk. If your pitch to buyers or investors is "we use AI," that line now subtracts value until you can show the signal underneath is non-replicable. So the move is to pull every data-partnership conversation forward, because the next model release resets buyer perception again and you want the ink dry first. The quiet problem Kawaja skips: owning unique signal and operationalizing it at scale are different jobs. Plenty of publishers sit on rich first-party data and can't plug it into a decisioning engine without a year of plumbing. For a non-specialist: having the data isn't the same as being able to use it, and the gap is where companies die.

The CFO. Underneath the thesis is a capital-allocation question. Money spent building on a frontier model is money you don't own the output of, and a repricing event every few months is exactly the kind of thing that wrecks a payback plan. The spend that survives scrutiny is spend on assets that stay on your balance sheet: data licenses, first-party collection, clean-room infrastructure. Those compound. A GPT wrapper does not. For a non-specialist: pay for things you keep, not for a thin layer on top of a tool your competitor can rent just as cheaply tomorrow.

The tensions. Kawaja and the Skeptic disagree on why the market froze: model uncertainty versus plain old rates and bad public comps. It matters, because if it's rates, the thaw comes from the macro, and "unique signal" is a sideshow. The Strategist and the Operator disagree on whether owning signal is enough: the Strategist says data provenance is the moat, the Operator says most data-rich companies can't actually turn that data into decisioning without heavy build. And everyone glosses the walled-garden risk the Skeptic names. If Google, Meta, and Amazon solve targeting natively for their buyers, "unique signal" in the open web is a shrinking prize no matter who owns it.

What it hinges on. Three beliefs. One, that signal scarcity is durable and not about to be papered over by a walled garden or a privacy-safe workaround. Two, that owning data and operationalizing it are close enough that the data-rich can actually capture the premium. Three, that the freeze is about AI defensibility and not mostly about the cost of money. The council leans toward Kawaja being directionally right on where value concentrates and wrong on why the market is frozen. Both can be true. Before committing capital to a "signal" story, verify the data is genuinely non-replicable (not just first-party, actually scarce) and that the company can plug it into decisioning without a multi-quarter rebuild.

Prediction: Between now and the end of Q2 2027, at least one data-rich, decisioning-capable public ad-tech or identity company (for example LiveRamp, PubMatic, or Criteo) will announce an acquisition of a smaller targeting-data or clean-room asset, while no venture-backed "AI for advertising" workflow tool announces a premium strategic sale in the same window.

Confidence: Medium. The direction of capital is clear; timing depends on when the deal market thaws at all.

Why: Kawaja is telling buyers, in public, that they should underwrite generic-AI ad-tech to zero and pay only for scarce data plus decisioning, and he is the banker those buyers call, so his taxonomy becomes the deal taxonomy. When a market reprices this way, the first checks go to acquirers bolting proprietary signal onto an existing decisioning engine at distressed prices, because that is the trade his own thesis endorses and the one with a defensible story to a board. The workflow tools don't sell at a premium in the same window because the whole point of the freeze is that nobody can price their moat, so they either wait or get acqui-hired cheap. The opposite outcome, a marquee premium exit for a GPT-wrapper ad tool, would require a buyer to bet against the exact uncertainty Kawaja just described to the entire industry.

Revisit by 2027-07-15: We're right if a data-rich decisioning or identity company announces a targeting-data or clean-room acquisition before end of Q2 2027 and no venture-backed AI-workflow ad tool announces a premium strategic sale in that window. We're wrong if an AI-workflow ad tool lands a premium exit, or if no signal-plus-decisioning deal gets announced at all.

One caveat worth owning: if the freeze is really about rates and credit, as the Skeptic argues, then the whole deal market stays shut and nobody announces anything. That would make the call wrong for a reason that has nothing to do with Kawaja's thesis, which is exactly why this sits at Medium and not High.