Turns out when you ask the only question that matters, nobody has an answer, and suddenly the entire sector started wobbling like a card table with one bad leg:

How do you know it works?

Not “how does the dashboard look.”
Not “did the platform attribute more conversions.”
Causally.
Provably.
Objectively.

The silence was… informative.

Over the past few months, I spoke with more than two dozen companies selling “AI optimization.”
Enterprise vendors, scrappy startups, the whole spectrum.

I asked four childishly simple questions:

This produced everything from PR panic to philosophical essays written in zero-gravity.
Translation: no one wanted to answer with math.

A senior executive put it bluntly to me — off the record — that most “AI” in this space is just deterministic machine learning in a trench coat. Linear regression. Decision trees. Reinforcement loops. Nothing wrong with that, but let’s not pretend it’s sorcery.

And the real kicker?

No one is letting a hallucinating LLM move billions in ad spend.

When I raised these questions publicly, something weird happened:

People told the truth.

Real, unvarnished, career-limiting truth.

Heather Carver said advertisers need independent proof, not platform storytelling.
Brian O’Kelley shrugged and pointed out this is the same confusion we’ve recycled since the exchange wars.
Javier Campos dropped the academic hammer: platforms are grading their own homework.
Eric Schwartz reminded everyone that “optimization” often means margin defense.
Oleg Korenfeld said the mess isn’t new — just rebranded.
Gareth Glaser noted that the transparency people want is mathematically impossible under MCP-driven systems.
Jon Bokor said advertisers cannot verify AI performance without complete transparency.
Eric Frenchman admitted he uses AI for things he doesn’t want to think about — then conceded he has no idea if any of it works.
Ayse Guvencer warned that most buyers can’t tell the difference between real AI and AI-shaped noise.
Patrick Berzai said AI still doesn’t understand people — just patterns.
Gregory Grímer asked why YouTube serves him feminine hygiene ads when his feed is 90% Jaguars and Roman Empire battles.
Judy Shapiro cut through it all: proving AI outcomes is a fool’s errand, and we’re replaying programmatic’s hype-disappointment cycle all over again.

Suddenly, the industry couldn’t stop accidentally confessing…

Then there was Kurt Donnel, CEO of Freestar — the only person who didn’t care about buzzwords at all.

He told me the real issue isn’t AI.

It’s leadership.

Companies keep launching AI projects with no defined problem, no baseline, no KPIs, and no clarity about what “good” looks like. It’s corporate cosplay dressed as innovation.

His fix is brutally simple:

And he’s right.
Transformation isn’t the goal.
Specific, provable improvement is.

He also said what every CFO knows but rarely says aloud:
If the real goal is reducing headcount or avoiding more hiring, leadership needs to say that, not hide behind “innovation decks.”

At Freestar, only the AI that teams actually want survives.
Not the AI consultants parachute in through a 68-slide pitch.

His metaphor?
AI should be the frosting — not the sprinkles.
Sprinkles look good.
Frosting adds value.

Underneath all this honesty is the one truth people only whisper:

AI in advertising is engineered to maximize platform revenue — not advertiser efficiency.

Platforms can’t voluntarily build tools that reduce spend or narrow targeting, because that reduces revenue and hits stock price.

So the “proof problem” isn’t a bug.
It’s a feature.

Opaque models protect margin.
AI “recommendations” push budgets in the right direction — just not the advertiser’s direction.
Agencies tolerate automation they can’t shut off because it keeps staffing costs down.
Everyone benefits from the illusion of intelligence.
No one benefits from transparency.

Advertisers cannot independently verify AI lift.
Platforms refuse transparency.
Agencies complain privately about automation they can’t override.
Targeting gets tighter and worse.
Creative fatigue accelerates.
Measurement collapses under its own contradictions.

Nothing is proven.
Everything is marketed.

This is the starting line.

Part II drops next — and it won’t be any gentler.

Want the full receipts, the transcripts, and the part everyone else is afraid to print?

👉 Upgrade to ADOTAT+
Because the truth is expensive — but the lies are even more costly.

Stay bold, stay curious, and know more than you did yesterday.

Terry Kawaja said the part no platform wants printed:
“AI isn’t democratizing anything. It’s consolidating power — and the trillion-dollar players already won.”

Investors told us which AI vendors to avoid — the ones that can’t show operating leverage, expand spend by default, or promise “optimization” without proof. DSPs without a real automation roadmap were at the top of the danger list.

Product managers admitted the things that never make it into webinars: AI suggestions boost platform margin first, defaults nudge budgets upward, and yes, manual overrides happen all the time — the kind you’d never know about from the glossy slides.

Measurement leads confessed the part everyone suspected: honest lift reports lose clients, synthetic benchmarks can conjure any number you’d like, and AI makes causal testing almost impossible.

This isn’t hype.
It’s what people say only when the cameras are off.

ADOTAT+ is the part they actually fear.y.