For the last several weeks, ADOTAT has been making an argument in pieces.

The argument was that AI optimization layers, across every major ad platform on earth, are systematically billing advertisers for sales that were already happening. We walked through the eBay 2014 study where 99.5 percent of paid clicks came back through organic search the moment paid search was turned off.

We walked through reward hacking, the four populations the algorithm actually finds, the four criteria a metric must meet to survive contact with two AI optimizers iterating against each other twenty-four hours a day. We walked through the override layer, the structural fix nobody wants to build because the structural fix slows the parade down.

That was the theory. This is the part where the theory stops being theoretical.

Because while we were writing all that, the receipts were piling up.

Two real Q1 2026 client campaigns ran through Viant AI, at major agencies, on behalf of brands you have heard of. A CPG breakfast staple with an app-based loyalty program. A digital-first bank running new-account acquisition.

Both watched the AI optimizer do exactly what we said it would do. In real time. On their dashboards.

With ROAS collapsing while the green numbers climbed.

The agencies caught it via internal holdout audits. Viant did not disclose it.

The four populations. The reward hacking drift. The proxy gaming the proxy. All of it, in production, on a brand budget, in the first quarter of this year.

We have the dashboards. We have the holdout numbers. We have the math. Parts Two and Three of this series are where they go.

But before we get there, you need to understand why we are starting with Viant Technology specifically, because Viant is not uniquely guilty. The Trade Desk does this. Google does this. Meta does this. Amazon does this.

Viant goes first because Viant wrote it down.

Here is a sentence. It comes from Viant Technology's FY 2025 10-K, the document the company files with the SEC under penalty of securities fraud liability. It lives in the Risk Factors section, which is where companies are legally required to tell shareholders, in writing, what could plausibly blow up the business.

"ViantAI could produce inaccurate, unexpected or unfavorable results or behaviors, and therefore poorly-designed or executed campaigns, sub-optimal ad placement, targeting, or bidding, or misleading insights for our customers."

Oh, and it gets better — admits that the AI will create “"Poorly-designed or executed campaigns."

Say it out loud. All of it.

Now here is what Viant tells its actual customers about the same product, on the ViantAI page, current as of right now: "unmatched precision."

And: "true deterministic data."

And: "Advertisers who use AI Bid Optimizer see up to 42% lower CPMs."

Same product.

Same fiscal year.

Two audiences.

The shareholders, who have lawyers and class-action mechanisms and the SEC enforcement division on speed dial, get the warning.

The customers, who have a procurement team and a DocuSign link, get the mood board.

I dialed Viant's sales team and pretended to be an interested advertiser. The bar for getting a sales rep to explain how the product works is "have a phone."

I asked the question every buyer asks. How does the AI actually decide what's working?

The rep told me, cheerfully, the AI will optimize to hit the easiest target possible.

He said it like it was a feature. Because in the frame Viant has built around its product, it is. Of course the system finds the cheapest path to a conversion. That is what optimization means. That is the whole pitch.

The 10-K, written by securities counsel, calls this misleading insights for our customers. The sales rep calls it the product. They are describing the same machine. And the machine, as we documented over the last several weeks, is doing exactly what reward hacking literature predicted it would do, finding the people who were going to buy anyway and billing for them.

This is the loop closing in real time. Theory, meet practice. Practice, meet a sales rep on a Tuesday.

Regular readers know this part. New readers, buckle up.

The entire industry is celebrating its way through a measurement crisis. ROAS numbers go up. CTV lift figures defy physics. Trade press writes "AI-powered outcomes" headlines from methodology papers that do not exist. CMOs get promoted. CFOs sign the checks. The dashboards are GREEN. The metrics are HITTING. The robots are CRUSHING IT.

The metrics are fake. Not "directionally questionable" fake. Fake fake. The optimization layer is finding people who were going to buy anyway, calling them incremental, and billing for it. We covered this in detail in the AI outcomes pieces. Now we are watching it happen on a specific company's specific clients.

Nike found out. Shifted budget into performance. Trusted the numbers. Stock dropped twenty-one percent in a single day. One hundred and fifty billion dollars in enterprise value, gone. That is the public number. The private number, distributed across every brand currently running platform-attributed campaigns, is much larger. You are paying it. You just cannot see it on any report you currently receive.

Viant has provided, in writing, signed by Viant officers, the cleanest paper trail in the industry.

The CEO criticized self-attribution and then bought a panel to bring self-attribution in-house. The COO admitted on the record there is "a lot of friction" in the core matching workflow. The SVP published a blog post warning about exactly the methodological failure the company's published case studies cannot rule out. The 10-K reads like a confession written under duress. The sales rep confirmed the whole thing on a phone call.

You don't need to follow the money. The money is sitting in the lobby waving.

July 2024. CEO Tim Vanderhook tells AdExchanger the central rot in digital advertising is self-attribution. "Every advertising platform measures its own performance today." He named the disease. Gold star.

April 2026. Viant pays $40 million to acquire TVision, an attention-measurement panel that had been selling its data independently to iSpot, VideoAmp, and Oracle.

Press release: TVision signals will now be exclusive to the Viant DSP.

The CEO who built his keynote around the self-attribution problem just bought the panel that was solving the self-attribution problem for everybody else, and locked it inside his own DSP.

This is what we in the industry call irony.

It is also what we in the industry call Tuesday.

Viant will tell you their measurement is "unmatched precision." Their SEC filing tells investors the AI may produce "misleading insights for our customers." Their CEO criticized self-attribution then acquired self-attribution. Their COO admitted "a lot of friction" in the core workflow. Their SVP warned about the exact methodological failure their case studies cannot rule out. Their sales rep cheerfully confirmed the whole thing on a phone call.

Every one of these people is telling the same story. The marketing pages are telling a different one. The dashboard is green. The parade is in full swing. The metric is fake.

You have been warned. You have been warned for free.

Part Two is where the autopsy starts. We brought receipts.

Part One is the warning shot. Parts Two and Three are the autopsy.

Part Two walks through the SEC disclosure gap line by l

ine, in a side-by-side table you can hand your CFO. And it walks through the two Q1 2026 client campaigns. The CPG loyalty program. The digital bank. With the dashboards. With the holdout audits. With the actual phase-by-phase math of the optimizer drifting from "loyal app scanners" and "fresh searchers" to bots, promo farms, and serial form-abandoners while ROAS collapsed and the green numbers kept climbing. This is the AI outcomes thesis colliding with reality on a real brand budget. We have the receipts.

Part Three is fourteen years with zero MRC accreditations. The MySpace user-registration data still on Viant's books in 2026, which is not a typo. The AlmondNet patent fight in Delaware, which AdExchanger called "the largest ID graph patent fight no one is talking about." The eight RFP questions that make Viant reps turn an interesting color. The contract language your legal team will thank you for. And the org-chart conversation about who inside your own company should own outcomes measurement, the answer to which is "not the team that buys the media."

REGISTER HERE