This is where the island getaways and thought-leadership cosplay finally meet the actual machinery.

Strip away the sponsored beach clubs and the "education summits" and what you're left with is something less theatrical and more dangerous: a measurement system that grades its own homework and calls it science.

Outcomes don't need a conspiracy to work. It only needs math, incentives, and your willingness to mistake "the dashboard turned green" for "we caused something real."

And when you read Viant's own internal documents—the sales memos and guidance materials meant for closers, not questioners—the shape of the trap becomes embarrassingly obvious.

Let's start with the most revealing decision: what the system optimizes for on day one.

The internal documents are explicit about this. Viant chose CPA and ROAS as the "lowest common denominator" metrics because—and this is their language—they're the clearest way to prove the AI "works" without having to "sell two things at once."

Translation: Incrementality is too hard to explain and too slow to prove, so we're starting with the metrics we know we can move.

If you were designing an autonomous ad product meant to prove superiority fast, you wouldn't lead with incrementality. You'd lead exactly where Viant led: CPA and ROAS.

Not because they're honest.

Because they're easy.

Easy to explain to a CFO who wants a number, not a philosophy seminar.

Easy to optimize by steering into known patterns: cheap supply, retargeting-adjacent behavior, generous attribution windows.

Easy to make look better than humans, especially when humans are constrained by approvals, budgets, and the need to explain their choices in rooms full of people who wish they weren't there.

Most importantly, they're metrics where "improvement" can be manufactured without lying. You don't have to create new demand. You just have to be better at claiming credit for demand that was already happening.

Steer toward inventory where pixels fire reliably. Weight toward audiences already on a purchase path. Let the attribution window breathe. Suddenly your CPA is $15 instead of $45, and everyone thinks the machine discovered gold when really it just discovered how to game the scoreboard.

This is not innovation. This is arbitrage disguised as intelligence.

And it works spectacularly well as long as no one asks the follow-up question: Did this cause anything, or did it just show up at the right time?