CMOtech Canada - Technology news for CMOs & marketing decision-makers
Canada
AI put campaign execution in the fast lane. Did we upgrade the brakes?

AI put campaign execution in the fast lane. Did we upgrade the brakes?

Mon, 16th Nov 2026 (Today)
Julia Larionova
JULIA LARIONOVA CMO PropellerAds

Removing friction from campaign execution has been a thorn in the side of marketing for decades. Professions have been built around it. Now, it's almost gone within a span of only a few years, thanks to AI. Thing is, all these manual handoffs and campaign setups did have another function. It's where judgement occurred.

The numbers suggest that the shift is already visible in everyday media buying. According to HubSpot's 2026 State of Marketing report, 19.2% of marketers are already using AI agents to automate marketing initiatives end-to-end, no intermediate click of approval, just outcome. Gartner says marketing leaders expect AI-driven automation to grow from 16% of marketing work in 2026 to 36% by 2028.

Execution is sprinting with record-breaking pace while judgment is still tying shoes.

That gap is where the real damage happens, and it isn't hypothetical.

Exhibit A: the metric that eats itself

Take Google's Performance Max. Left unconstrained, its bidding algorithm will cheerfully bid on a brand's own branded search terms, traffic that would have converted anyway, for a fraction of the cost, and book it as a win. PMax can capture demand that is already close to conversion, including branded searches, making attributed ROAS look stronger without necessarily proving the same level of incremental value.  

Optmyzr found Search/PMax keyword overlap in 91.45% of the 503 accounts it studied. Some of that overlap may have legitimate eligibility or budget explanations. But the broader problem remains: an optimizer rewarded for attributed ROAS has no philosophical objection to harvesting demand the brand already created. And don't mistake it for a misfire. That's the default behavior of an algorithm doing exactly what it was told. Unless you teach it the algorithm sees no difference between creating value and finding the easiest place to claim it. 

Exhibit B: AI that skipped Anatomy 101

Meta's Advantage+ Creative suite auto-enrolls advertisers into AI-driven edits. So they crop away, add backgrounds, swap images, often without anyone opting in. REI found itself running an Instagram ad depicting a bike with two sets of handlebars. Another advertiser's fitness ad showed a model with a leg bent at an anatomically impossible angle. Meta sends no alert when the tool modifies your creative. REI said it hadn't asked for the alteration. Meta's response pointed back to its terms: advertisers are responsible for reviewing AI outputs.

Think about that division of labor for a second. The platform gets the autonomy. The brand keeps the responsibility.

Multiply either of these across every campaign an autonomous system now touches, and "human in the loop" stops meaning anything useful.

"Human in the loop": is it a phrase or a plan?

Here's the problem with keeping "a person involved": there's no single decision where oversight quietly becomes automation running unsupervised. It's the sand-dune problem: when does one more grain turn a handful into a beach? How many of them do? A media buyer cannot review every bid adjustment. Nobody's rubber-stamping every AI-generated crop. Adding an approval step everywhere just rebuilds the bottleneck AI was supposed to remove.

So the real question isn't whether to keep a human in the loop. It's which decisions actually deserve one.

Three things need defining before any team hands over more autonomy:

  1. The objective an agent is actually optimizing, and how far that metric sits from the outcome the business cares about. Revenue and registrations are not the same mandate, even on identical inventory.
  2. The evidence threshold before a system acts. A performance shift based on twenty conversions shouldn't carry the same weight as one based on twenty thousand.
  3. The escalation line - which decisions are cheap and reversible, and which ones move real budget or go public on the brand's behalf, and therefore need a human who can be held accountable, not just "in the loop."

The scoreboard problem

McKinsey found 90% of CMOs are experimenting with AI, and fewer than 10% have scaled it or captured real value. Most of what gets measured - assets generated, campaigns launched, hours saved - rewards activity, and never judgment. None of it tells you whether the PMax problem or the REI problem is happening inside your own accounts under your radar.

Campaign execution is fast now - fast enough that a bad call gets expensive before a human even knows there was a call to make. The organizations that get this right won't be the ones running the most AI-driven campaigns. They'll be the ones who can say, specifically, what their agents are and aren't allowed to decide on their own - and who finds out first when one of them gets it wrong.