Red Flag #7

Every flat or declining month is blamed on a "learning period"

There is a real thing in Google Ads called a learning period. When you make a meaningful change — a new bidding strategy, a significant budget shift, a structural campaign change — the algorithm needs roughly two to three weeks of data to recalibrate. During that window, performance can be unstable. After it, performance should stabilise and you should be able to read the new normal. This is real, and on a healthy account it gets cited maybe once or twice a quarter.

The red flag is when “we’re still in the learning period” becomes the standing answer for every month that doesn’t go the right way. It is one of the most reliable signals on this list because it requires no looking at the account to detect. You just have to count how many months in a row you’ve heard it.

Why agencies do it

It is the most defensible-sounding rhetorical shield in the whole vocabulary.

A learning period is a real concept, documented by Google, with real mechanics. It cannot be dismissed by a non-practitioner without sounding like the non-practitioner doesn’t understand the platform. So an account manager who has nothing useful to say about why a month went badly can reach for “learning period” the way a politician reaches for “it’s complicated” — technically correct, almost always insufficient, and effective at ending the conversation.

The mechanic underneath is usually one of three things. Either the agency is making frequent small changes that keep retriggering the learning state (which is itself a process problem), or the agency made a single change two months ago and is still claiming we’re calibrating (which is a credibility problem), or the agency has made no change at all and is using “learning” as a generic term for “we don’t want to commit to why this happened” (which is the worst version).

What it looks like in your report or account

The phrase “learning period” appears in two or more consecutive monthly meetings.
When you ask, “Learning from what change, made on what date?” you don’t get a specific date or a specific change.
The campaign change history (visible in Google Ads under Tools > Change History) shows either constant micro-edits or no meaningful changes for the past 30+ days.
Performance stays flat or declines but the explanation doesn’t evolve. Three months in a row of “still calibrating” is not three different explanations.
Smart Bidding strategies (Maximise Conversions, Target CPA, Target ROAS) are being changed every four to six weeks, which keeps the account perpetually in the learning state by design.

What to ask your agency

Two questions, in this exact order. The order is the trap.

First: “What specific change triggered the current learning period, and on what date?”

Second, regardless of what they answer: “What does the trailing-90-day trend look like? If we’re still calibrating from a change made over three weeks ago, that’s longer than the platform documents. What’s actually happening?”

Good answer

“The most recent meaningful change was a switch from Maximise Conversions to Target CPA on the consideration campaign on April 14th. We’re fifteen days in. The first two weeks looked unstable, which is normal. Last week stabilised at a CPA of $284, which is 8% over our $260 target. We’ll let it run another seven to ten days to confirm and then either tighten the target or roll back. We’re not making other changes during that window so the data stays clean.”

Bad answer

“Smart Bidding takes time to find its footing, especially in a competitive market. We’re continuing to monitor and we expect performance to normalise as the algorithm finishes calibrating. Patience is critical with machine learning systems — pulling levers too aggressively just resets the learning.”

What it means if you get the bad answer

It means the agency is using a real Google Ads concept as a fog of jargon, in the hope that you don’t know enough to notice. It is the verbal equivalent of slowing down at a yellow light forever. Calling it out by name — “a learning period documents at two to three weeks; we’re past that, what’s the actual issue?” — collapses the rhetoric. Either you get a real answer or you get the next layer of fog, and the next layer of fog tells you everything you need to know about whether to keep going.

In my own auditing work, this is the single most common phrase I see used as cover. When I get into the account and pull change history, the “learning” that’s allegedly happening is one of two things. It’s either nothing — no real changes for weeks — or it’s churn, where the agency keeps tweaking small settings that keep restarting the algorithm. Both explanations are real. Neither is the explanation the client got.