LinkedIn
“Don’t touch it, it’s still learning” is how known waste survives another two weeks in a Google Ads account. I used to give clients that line. It sounded responsible, bought me a quiet fortnight, and left the account spending while obvious problems went unfixed.
Here is the claim: the learning phase is a status label, not a quarantine. Google defines Learning as a bid-strategy status with four listed causes: a new or reactivated strategy, a setting change, or a composition change such as adding or removing campaigns, ad groups, or keywords. The label tells you bidding is adjusting. It does not tell you to leave broken tracking, the wrong conversion goal, or junk search terms alone.
The two-week rule confuses a status with an instruction
Calibration depends on conversion volume and the time between a click and a conversion, not just the number of days on a calendar. One learning-phase explanation describes a period of up to around 50 conversion events or three conversion cycles. It also notes that Manual CPC has no learning period and that changing a tCPA or tROAS target does not erase everything the model has learned. None of that adds up to a universal two-week ban on account management.
The popular advice goes like this: after a launch or major edit, go hands-off for one to two weeks. Make one big optimization in week three, then settle into weekly check-ins. I preached that version myself. People recite the usual list of disruptive changes: big budget moves, targeting changes, long pauses, added or removed ad groups, and strategy switches. Touch one, restart the clock. Better not touch anything.
That last sentence is where the money leaks out. A structural change may be worth planning carefully. A broken conversion action is not something to admire from a safe distance. “Avoid unnecessary resets” is sensible advice; “avoid necessary fixes” is not.
Low-volume accounts cannot wait their way into volume
Say you spend $20k a month on a B2B account that closes 18 conversions a month. Guidance that calls for roughly 30 to 50 conversion events in a short window before bidding stabilizes creates an awkward problem: that account cannot produce the suggested volume just because its manager leaves it alone for fourteen days. A cautious test budget may mean a campaign stays in Learning long after the calendar reminder says to check back.
Two weeks pass, then four. The status still says Learning, or perhaps Learning Limited. The agency calls that patience. I call it paying full price for an account nobody is willing to manage.
This is familiar territory for smaller lead-gen and ecommerce campaigns. Budget guidance flags 15 to 30 conversions a month per campaign as a difficult range for Maximize Conversions or tCPA. Another guide puts 30 conversions as a minimum and 50-plus as a stronger target for tCPA, with lower figures for Maximize Conversions. Those numbers are rules of thumb, not a switch that flips bidding from useless to brilliant. They do make one point hard to dodge: waiting does not create conversion volume.
Worse, waiting does not repair bad inputs. If tracking counts the same conversion twice, bids respond to inflated results. If the goal is a raw form fill but the business needs qualified pipeline, the account can get better at buying the wrong leads. If broad match keeps finding DIY searches instead of buyers, another quiet week gives those searches more budget. The model can settle down while the business problem stays exactly where it was. A stable CPA is not much comfort when it measures the wrong action.
The meter runs throughout. At $20k a month, an illustrative 18% of spend on irrelevant search terms is about $900 in a week. You do not recover that money by pointing to a status column. Let the same waste continue for a month and it is about $3,600. A learning label never justifies leaving known waste in the account.
I used to tell clients to wait. I was wrong.
My old rule was easy to explain: launch on Maximize Conversions, set a loose tCPA, and leave it alone for fourteen days. It also protected me. Any bad week had a ready answer: still learning. The client got a status update when what they needed was someone to look at what the account was buying.
Take a home services account at roughly $20k a month and 11 conversions a month. Split those conversions across three near-identical campaigns. Let broad match buy drain-cleaning DIY queries, then count every phone click over ten seconds as a lead. The account may sit in Learning Limited for weeks, but the label is not the cause of the problem. The campaigns split a small signal three ways, the searches include people unlikely to book, and the conversion goal rewards clicks that may never become jobs.
I would once have waited for that account to clear learning before making a serious change. I was wrong. It needs fewer competing campaigns, a goal tied to qualified calls, and negative keywords for the junk terms. Those edits may disturb bidding. Good. The account needs a better lesson, not more time with the old one.
Ask whether the change is worth a reset
Forget the calendar. The question is “Is this change worth a reset?” If it gives bidding a more useful signal or stops material waste, make it and accept the possibility that Learning appears again. If it merely rearranges the account for neatness, batch it. Cause first, label second.
Some fixes should not wait for a quieter status:
- Broken or doubled tracking. If purchases, calls, or form fills are missing or overcounted, the account is bidding against bad information. Fix the measurement before asking whether performance has improved.
- The wrong conversion goal. Raw form fills are not qualified pipeline. Use the action closest to revenue that you can feed consistently; otherwise, better bidding can still buy worse business.
- Junk search terms spending real money. If career or DIY queries are taking a meaningful share of a B2B budget, add negatives when you find them. Negative keywords are generally treated as a safe adjustment that need not reset learning. Waiting for a ceremonial day fourteen serves nobody.
Other changes deserve coordination, not paralysis. Structural splits, audience expansions, and large budget jumps can alter what a campaign is trying to learn. If you need to scale, smaller budget increments are easier to judge than one sudden jump. If three campaigns share 30 conversions a month, ask whether the separation earns its keep. You may not have three distinct opportunities. You may have one thin signal cut into three thinner pieces.
For an account stuck in Learning Limited, changing what you optimize toward, consolidating fragmented campaigns, or pooling campaigns through portfolio bidding can be more useful than another round of waiting. These are not cosmetic edits, and some may trigger fresh learning. That is the cost. The alternative is to keep paying for a structure that cannot gather enough signal to support the way you want to bid.
Then there are the ordinary chores that should never have been trapped behind the quarantine rule. Adding negatives, making modest bid adjustments, and testing creative or ad copy are commonly treated as lower-risk changes. Their precise effect depends on the change and the bidding setup, but the point stands: a Learning label is not a reason to stop checking search terms or improving an ad. Judge an edit by what it changes, not by whether someone can blame it for a label.
Put a 90-day number where the excuse used to be
A written target makes the conversation harder to dodge. Ask for the CPA on qualified conversions, the volume you expect, and the date you will judge against. Then ask what gets fixed each week. If the only answer is “we cannot say; it is still learning,” you have not been given a plan. You have been given a status report.
The target also forces a useful distinction. A campaign can spend its way toward more form fills while moving no closer to qualified calls or closed-won revenue. Without an agreed outcome, every busy-looking adjustment can pass for progress and every idle week can pass for restraint. With one, you can ask whether the changes improved the number the business actually needs. For more on the status itself, read how long the learning phase lasts and what resets it and the week-one checklist of what to fix first.
That is why I want the target before the spend. At groas, paid work runs as 90-day sprints with the number written down before launch: 40% more sales and leads at a cost per sale no higher than day zero, or the team keeps working for free. Put that commitment on page one and “don’t touch it” stops sounding like a management strategy. Search-term waste gets cut, tracking gets checked, and fragmented campaigns have to justify their existence. The action log shows what happened instead of asking you to trust that somebody watched a dashboard. Continuous execution leaves no room for a two-week alibi.
There is an account where leaving bidding alone for a stretch makes sense: high volume, clean tracking, a consolidated campaign, and enough conversions for the strategy to find its footing quickly. That is not the $20k account with 18 conversions a month. It is not the home services account measuring phone clicks as jobs. Using the same hands-off rule for all three is not discipline. It is a convenient way to avoid deciding which changes matter.
Keep honoring the quarantine and the bill is straightforward. At the illustrative $20k monthly spend above, 18% on irrelevant terms is about $3,600 a month for as long as nobody cuts it. The account may also keep teaching bidding that those searches, or those easy form fills, are valuable. When performance finally looks steady, you still have to ask what it has steadied around. You did not protect learning. You funded the wrong lesson.
So stop asking whether the campaign is learning. Ask whether the change is worth a reset. Fix broken tracking, the wrong goal, and expensive junk terms today. Plan disruptive structural changes for the signal they can improve, not for the color of a status label. Demand the CPA, the qualified volume, and the date in writing; judge every week against them. Anything else is what the deck calls caution. I call paying someone to watch your budget spend itself.
Send whoever runs your account this line on Monday: show me CPA on qualified conversions for the last 14 days, the three changes you made while it said Learning, and the 90-day target in writing. If the reply is another lecture on patience, you have your answer. Accounts do not improve when the label clears. They improve when someone fixes the signal underneath it.