“Don’t touch it for two weeks” is expensive advice for a new Google Ads campaign. At $20k a month, you are spending roughly $650 a day while search terms, landing pages, and conversion tracking are at their least tested.

Let that run untouched for 14 days and you have spent more than $9,000. Some of those clicks may be useful; some may be teaching Smart Bidding from bad data. I used to tell clients to wait. I was wrong. The learning phase is not a waiting room, and most urgent fixes do not require you to sit through it.

The fear behind the advice is real. Early Smart Bidding looks shaky. CPCs swing, conversion volume stutters, and the status column flashes Learning in orange like a warning light. The folk theory says any edit restarts the clock, so hands off or you will pay for another two weeks of bad bidding. I believed it because it gave clients a clean story: be patient, let the machine learn.

But patience has a price. New campaigns without negatives can waste 15 to 30% of budget on bad traffic in the first week. Think free, jobs, DIY, and reddit queries that should never have matched in the first place. On a $20k-a-month account, that is roughly $1,400 to $2,700 in seven days. Bad traffic does not speed up calibration. It gives the bidder bad examples.

Stopwatch melting into dollar bills over a laptop showing Google Ads

The learning phase needs conversions, not 14 days of silence

Smart Bidding is not warming up like an engine. It is building a prediction for each auction: how likely is this click to produce the conversion you told it to chase? To make that guess, it needs examples. Google puts the calibration window at up to around 50 conversion events or about three conversion cycles, faster when the account has history. Notice what is missing: a promise that everything settles on day 14. Volume sets the timetable, not the calendar.

That matters because Google capped the visible Learning label at seven days maximum in October 2026. The label can clear while performance is still settling. If a campaign gets four conversions a week, the end of that visible status is not proof the bidder has learned enough. Waiting another seven days by habit does not solve the volume problem, either.

The October 2026 rewrite narrowed the trigger list. Learning starts for a new or reactivated bid strategy, a change to the campaigns, ad groups, or keywords in a strategy, or a significant change to goals or conversion actions. Google’s help page describes the other side of that line: budget, ROI-target, and targeting updates take effect in the next auction without triggering Learning status, while creative edits retain history.

That distinction does not mean every edit is wise. It means “an edit happened” is a poor diagnosis. First ask whether you changed what the bidder optimises for, what inventory its strategy covers, or the examples it receives. Then decide whether the change can wait. A broken conversion tag cannot. A new campaign structure usually can.

Read the trigger list again before applying the old 20% rule. There is no percentage threshold for budget changes in Google’s pages on learning or bid statuses. The rule was folklore we used to sound careful. Google product manager Tal Akabas put the mechanism plainly: on an established campaign with steady history, target or budget changes do not restart learning because bids can scale from existing auction data. A new campaign has less history to lean on. That is a reason to avoid wild swings, not a reason to ignore a broken page.

Change the strategy deliberately. Fix the waste immediately.

I separate launch edits by what they change, not by which day appears on the calendar.

A bid-strategy switch changes the job. Moving from Max Clicks to tCPA, or from tROAS to Max Conversion Value, tells the bidder to pursue a different outcome. A composition change alters the auction pool the strategy controls: campaigns, ad groups, or keywords added to or removed from it. A significant goal change alters the definition of success. Swapping the primary conversion action, changing its counting, or rewriting value rules belongs in that last category. Make those decisions deliberately and batch the changes you can plan. Do not rebuild the account every Friday because the first few days look ugly.

Big budget and target moves sit in a different category. Under the current definition, they do not trigger the Learning label or wipe historical data. But doubling spend overnight on a fresh campaign can still whip bidding around. The mechanism is different; the volatility is still yours to pay for. Set a sensible daily budget and realistic target at launch, then adjust in batched steps instead of daily yanks. Protect stability without confusing it for a reset.

A control panel with three red levers labeled strategy, structure and conversions, and a row of grayed-out dials for budget and ads

Now look at the work the two-week rule delays. Search term mining belongs near the top. Adding negatives does not put a campaign back into Learning, and it stops irrelevant queries from consuming the next day’s budget. Weekly mining and negative additions can improve performance without a reset. I would not wait a week when the first terms are already telling me the campaign matches jobs instead of buyers.

Creative deserves the same attention. Fix a typo, replace an off-brand headline, and correct an ad that sends people to the wrong offer. Creative fixes are safe during learning; the edit does not erase historical data. If fixing a problem means adding or removing keywords or otherwise changing what a bid strategy controls, treat that separately. Batch a non-urgent restructure. Stop clearly irrelevant spend when you find it, even if that intervention has a learning cost. Paying for known-bad clicks is not a strategy for protecting a status label.

The two problems people are most scared to touch are the ones I check first: tracking and pages. If a tag double-fires, misses the thank-you page, or imports offline conversions with the wrong date, waiting teaches Smart Bidding from the wrong record. Fix the measurement problem immediately, even if correcting a conversion action also changes the learning state. That is the tradeoff. You cannot protect a model’s learning by feeding it false answers.

Same with a landing page that loads in six seconds, throws a consent wall, or 404s on mobile. Fix it or stop sending traffic there. A bidder cannot turn a broken form into a customer. Clean data and a working page matter more than an uninterrupted Learning label.

“Strategic patience” can mean nobody looked

I used to tell clients to wait because waiting protected me, not them. When you manage six accounts and bill by the hour, “leave it alone for two weeks” is convenient. It buys time and gives poor early numbers an excuse. If the client swapped a landing page without telling you, which happened to me twice, you can blame learning instead of slow QA. What the deck calls strategic patience, I now call not looking at the search terms.

What broke the habit was the pattern. Day-three search terms full of free and jobs. Day-five conversion tag firing twice. Day-nine ad with a typo the client approved at midnight. None of those problems fixes itself, and the account does not pause spending while everyone waits for a calendar date. I started mining terms daily and fixing tags on sight. Accounts settled faster, not slower.

That does not make every day-three number decisive. An early CPA jump can be noise; an irrelevant search term is a fact about a click you bought. A conversion tag firing twice is a measurement error, not a performance trend. Put those in different buckets. Investigate the signal before you overhaul the strategy, but do not use uncertain performance as an excuse to leave certain defects live. The bidder learns from the traffic and conversion data you give it.

Low-volume accounts still need attention, not daily reinvention

Here is who should skip the aggressive version of this routine: an account driving fewer than about 30 conversions a month. With four or five conversions a week, Smart Bidding may not get the roughly 50 conversions or three conversion cycles it needs to settle quickly. Every performance judgment then rests on thin data. Daily search term mining still helps, and a broken tag still needs fixing. Daily strategy changes do not.

For those accounts, consolidate campaigns, bid to a higher-funnel action that actually fires, and resist the urge to swap strategies every Friday. That is not the same as leaving the account alone. It is choosing changes you can defend. If a query is plainly irrelevant, exclude it. If CPA moved after two conversions, do not rebuild the campaign to explain it.

For accounts with more volume, the active routine beats the waiting routine. I go deeper on reset timing and spend protection in how long the learning phase actually takes. For the month after launch, use the week-by-week schedule for the first four weeks. The principle behind both is simple: touch the account, just touch the right parts.

The first two weeks: check the account, not just the status

Keep this next to the account. Work through the checks in order, repeat the search-term and page checks as new traffic arrives, and batch structural decisions where possible. These are jobs, not permission to make 13 strategy changes.

  1. Day 1: Test the primary conversion. Confirm it fires once per real customer and shows as Recording; duplicate tags or the wrong counting setting corrupt the signal.
  2. Day 2: Pull the first 48 hours of search terms. Add relevant campaign negatives for queries such as free, jobs, careers, DIY, tutorial, reddit, course, and template; check intent before excluding a term.
  3. Flag keywords with spend and no relevance. Stop clearly irrelevant spend; batch any broader keyword restructure rather than making one change after another.
  4. Read every live ad. Fix typos, wrong offers, and landing-page mismatches while the creative history stays intact.
  5. Load the landing page on mobile with throttled 4G. If it stalls past three seconds or the form breaks, stop sending that ad there until the page is fixed.
  6. Confirm location, language, and ad schedule against the brief. Wrong geography can burn budget faster than wrong copy.
  7. Check Change History. Find out who touched the account since launch before you blame the bidder for a change someone made.
  8. Mine search terms again. Note repeat winners for a planned keyword change rather than adding them one at a time without considering the strategy’s composition.
  9. Check bid status. Learning alone is not a defect; investigate Limited or Misconfigured before you decide the account merely needs more time.
  10. Adjust tCPA or tROAS if spend is capped or starved. Make one considered move, then give the result room to show up instead of nudging it daily.
  11. Add the next round of negatives from days four to seven. Phrase match keeps exploring after the first cleanup.
  12. Compare Google Ads conversions with GA4 and your backend. Small gaps are normal; a large gap deserves a tracking check before a CPA verdict.
  13. Reconcile imported conversions and date ranges before judging CPA. Offline imports stamped to the click date can make recent days look empty; skip this check and you may pause a winner because attribution arrived late.

That routine is the busywork I stopped doing by hand. I run launches through groas now because this is what it does well: mine search terms daily, add negatives before waste compounds, check tags and pages while spend is still small, and log each action with its reasoning. A human strategist still sets the targets and guardrails. The account gets attention on day three instead of day fourteen.

Leave a new campaign alone for two weeks and you do not buy cleaner learning. You risk spending more than $9,000 before catching clicks you already knew were worthless, a page that cannot convert, or a tag that tells the bidder the wrong story. The bill keeps running while you call it patience.