Do you need a written CPA target before launching a new Google Ads campaign?

Yes. The author used to think a vague goal was fine for the first two weeks while data gathered, and now considers that wrong. A written target, such as 25 qualified calls a month at $92 CPA, does not stabilize the campaign by itself, but it gives you something to judge once the early noise settles and stops every bad day from becoming a debate.

Should you change budgets or bids while a new Google Ads campaign is still in the learning phase?

No. Changing bids, budgets, targeting or creative during learning can disrupt learning and reduce serving. The author advises small corrections only, such as adding negatives from the Search Terms report or fixing ad copy that mismatches the page, and logging them, rather than rewriting the plan because of one or two calls.

How long does the Google Ads learning phase usually last and how many conversions do you need?

The practitioner rule given is one week to one month to exit learning, with 15 conversions in 30 days as a minimum and 30 to 50 a month for best results. In the example, 11 conversions in the first seven days on $840 of spend was enough to keep watching but not enough to declare victory.

How can an unlogged landing page change hurt a Google Ads campaign?

When the client pushed a new site template, the service pages became slower, moved the form two scrolls down below an auto-playing video, and no longer matched the ad's promise. Quality Score on the money keywords fell from 7 to 4, average CPC rose from $4.10 to $5.52, and impression share lost to rank climbed nine points. Google Ads Change History only records changes in the ads account, so the page swap was invisible there.

Why did reported conversions drop by a third overnight after fixing the conversion tracking?

Because the count had been inflated. The site redesign left a button-click event marked Primary, so the account was counting button clicks as conversions for two days. When Primary was stripped back to the qualified-call event, reported conversions dropped by about a third, but no calls had been lost; the metric simply stopped treating button clicks as calls.

What landing page fixes helped Quality Score and CPC recover after the redesign?

The service pages were rolled back to the old copy, and the form was rebuilt on a staging link before going live: the headline matched the ad, the phone number sat in the header, the form sat above the fold, and the video played on click instead of auto-playing. Quality Score climbed from 4 to 6 in six days and back to 7 the following week, and CPC fell from $5.52 to $4.35.

What should you write down before spending a dollar on a new Google Ads campaign?

Three lines beside the build tab: the target (for example, 25 qualified calls a month at $92 CPA), the definition of a qualified call, and a change rule giving 48 hours' notice before a page publish, with every site edit recorded in the same shared log as the ads edits. A log entry takes about twenty seconds; discovering an unlogged page at 1am takes much longer.

Why use a 90-day clock instead of a rolling monthly target for a Google Ads campaign?

A 90-day clock sets a fixed schedule: thirty days to buy clean signal, then sixty days to judge CPA against the written target. A rolling monthly review tends to become a shrug, while a defined schedule tells you when to check the account, though it only protects the work if someone owns the target and everyone who touches the funnel records their changes.

Day 11: The Landing Page That Undid a Google Ads Launch

The search campaign had looked fine on Friday. Now it was 1am, the coffee beside my laptop was cold, and two of its best keywords had gone from Quality Score 7 to 4.

This is a composite of launches that broke in much the same way. In this version, it is Monday night turning into Tuesday in week two. I am at a kitchen table, working backward through an account I thought I knew. Average CPC has jumped. Impression share has slipped. The change history offers no obvious reason. Somewhere between the ad and the customer, something has changed.

Scene 1: A clean build with one missing page

On launch day, the plan lived in a spreadsheet with three tabs: build, negatives, checks. Each service had its own ad group, with two responsive search ads and a landing page matched to that service rather than the homepage. I typed out the exact and phrase match structure by hand. The starter negative list held about 80 terms around jobs, free, DIY and how-to queries, because that is where a home services budget can go to die in week one. Budget was $120 a day. I used Maximize Clicks with a cap for the first few days to buy data without letting the auction run wild.

It was familiar work: slow, manual and worth doing. I could point to the ad for each service, the page it would send someone to, and the queries we did not want to pay for. Anyone looking at the build tab would have seen a launch plan. I did.

The missing page was the one that said what success meant and who could change the funnel. The client wanted more good calls. I took that to mean a lower cost per acquisition and moved on. We had no written CPA target, no shared definition of a qualified call and no agreement about changes to the landing pages. The spreadsheet recorded what I built. It did not record what someone else might replace.

I used to tell clients a vague goal was fine for the first two weeks while we gathered data. I was wrong. A number does not make a new campaign stable, but it does give you something to judge once the early noise settles. Without one, every bad day becomes a debate about whether the account is learning or leaking. On day zero, that debate still felt theoretical.

Scene 2: Days 3–7, when doing less felt harder

By day three, the status column showed Learning. Clicks came in lumps. One morning brought three conversions before 9am; then we went two days without one. Google advises allowing for fluctuations and waiting until learning is over to judge performance. Sensible advice. It is also hard to deliver calmly when a client can watch $120 leave the account every day.

The practitioner rule I use is one week to one month to exit learning, with 15 conversions in 30 days as a minimum and 30 to 50 a month for best results. We had 11 conversions in the first seven days on $840 of spend. That was enough to keep watching, not enough to declare victory. I had no agreed definition of a good call to put beside the conversion count, either. The dashboard could tell me what it had counted. It could not settle what the client meant by good.

The urge to intervene arrived on schedule. The client asked whether we should raise budget on the ad group with two good calls and pause the one with none. I said no. Changing bids, budgets, targeting or creative in this stretch can disrupt learning and reduce serving, and two calls were not a reason to rewrite the plan.

I did not sit completely still. I mined the Search Terms report and added 23 negatives. I also fixed two ads whose headlines promised same-day service that the pages did not mention. Those were edits, and I logged them. They were small corrections to a visible mismatch, not a new bidding strategy disguised as a Tuesday afternoon idea. If you want the tidy calendar version, the week-by-week plan for the first four weeks lays it out. From inside an account, week one feels less like a calendar and more like deciding, repeatedly, whether the thing bothering you is worth touching.

By Friday, the two keywords carrying half our conversions still showed Quality Score 7. I closed the laptop with the build intact, the budget intact and the pages, as far as I knew, intact.

Scene 3: Day 11, the account looks hungover

Monday morning, average CPC on the money keywords had climbed from $4.10 to $5.52. Impression share lost to rank was up nine points. The two keywords that had shown Quality Score 7 on Friday now showed 4. That is an unpleasant set of numbers to meet before coffee, even when you know a new campaign can wobble.

Quality Score is a 1-to-10 keyword-level diagnostic built from expected CTR, ad relevance and landing page experience. It is not itself the number that buys a click. But a falling score gave me somewhere to investigate: the relationship among the query, the ad and the page. A weaker landing page experience can matter to the auction. The question was why that experience had changed when the ads had not.

That night I went through the usual suspects. Bids were untouched since day four. Budget was untouched. The Search Terms report showed no new junk cluster. Ad strength still said Good. I opened Change History, filtered it to the last 14 days and scrolled past my negative additions and the two headline fixes. No bid change explained the jump. No budget change explained it. Nothing there showed a new page.

The absence was useful, but only up to a point. Change History could show me what had happened in the ads account. It could not tell me whether someone had changed the destination after a click. I had been treating the ad and landing page as one piece of work; our records treated them as two separate worlds. At 1am, that distinction stopped being administrative.

I opened the ad preview.

Scene 4: The page nobody logged

The click landed on a page I did not recognize. It had a new hero image and a new headline. The lead form sat two scrolls down, below a brand video that auto-played. I had matched the ad to a service page on day zero; this page gave me a different first answer to the same search. I went back to the spreadsheet to check the destination, then opened the page again. The URL was not the mystery. The page behind it was.

I called the client and got the marketing manager on speaker. They had pushed a new site template Thursday night. Marketing wanted consistent branding across the pages. Nobody had told me. There was no email, Slack thread or shared document entry to find, because no one had made one. The page was prettier. It was also slower, placed the form farther from view and said almost nothing about the service named in the ad.

Same keyword. Same ad. Different promise on arrival.

That was a plausible reason to look hard at landing page experience and the rise in CPC. It was not proof that the page swap had, by itself, reset the bid strategy’s Learning status; Google describes that status in terms of bid-strategy, setting and composition changes. I did not need to pretend the status label explained everything. The customer had arrived somewhere different, and the campaign was now collecting clicks and conversions against that different experience. History from the old page could no longer tell the whole story.

A changed service landing page open on a laptop beside a cold cup of coffee

The page was not the only thing the template had changed. It had also broken the thank-you page trigger our conversions depended on. For two days, the account counted button clicks as conversions because a button-click event had been left marked Primary during the redesign. A click on a button and a qualified call are not interchangeable outcomes, however neatly they appear in the same column. The first 30 days of conversion data are a poor time to teach an automated strategy that the easier event is the one to pursue.

I stripped Primary back to the qualified-call event and retested the conversion path with a live submission. Reported conversions dropped by a third overnight. That number looked awful if you read it without the change log. Read alongside the tracking fix, it meant the count was honest again. We had not lost a third of the calls in one night; we had stopped calling button clicks calls.

I had gone into the week thinking I was protecting a young campaign from unnecessary edits. Someone else had edited the page underneath it, and the measurement had shifted too. The ads dashboard had recorded neither decision in a way that told the story.

Scene 5: Weeks 3–4, getting back to the original promise

We rolled the service pages back to the old copy on a Friday afternoon. I rebuilt the form placement on a staging link before it went live: the headline matched the ad, the phone number sat in the header, the form was above the fold and the video played on click instead of starting on its own. These were not glamorous changes. They made it possible for someone who clicked a service ad to see that service without hunting for it.

Quality Score climbed from 4 to 6 in six days, then to 7 the following week. CPC fell from $5.52 to $4.35. I would not credit one line of copy for every movement in an auction. I could, however, see the page, the tracking and the account moving back toward the setup we had meant to run. The practical advice in guides to shortening learning is less glamorous than the phrase sounds: make fewer disruptive edits, keep tracking stable and give clean conversions time to accumulate.

Volume returned more slowly than price. We logged 19 qualified calls in week three and 24 in week four, finally reaching the 30-to-50-a-month band I wanted to see before trusting a pattern. I checked Search Terms every Monday and Thursday, added another 41 negatives, and left budget and targets alone for 16 straight days. After the scramble around the template and the tracking, that stillness felt suspicious. It also gave me a run of data I could read without first asking what had changed behind it.

By day 28, CPA sat 22% below day 10. I had not found a clever setting. We had repaired the page, corrected what counted as a conversion and stopped making the account answer for unrecorded changes. Recovery had taken time because we had asked the campaign to work through two different versions of the same funnel in its first month.

Coda: The sheet I would write before spending a dollar

If I could redo day zero, I would put three lines in writing beside the build tab:

Target: 25 qualified calls a month at $92 CPA.
Definition: A qualified call comes from a homeowner in the service area asking for a date.
Change rule: Give me 48 hours’ notice before a page publish, and record every site edit in the same shared log as the ads edits.

Those lines would not have prevented every bad day. They would have told both teams which days were bad, what we meant by a good result and where to look when a number moved. Date, person, change, reason. It takes about twenty seconds to write an entry. It takes much longer to discover an unlogged page at 1am.

I would put that page inside a 90-day clock, too, rather than a rolling monthly shrug: thirty days to buy clean signal, then sixty to see what CPA we can judge against the written target. That is the premise of a 90-day sprint with the number in writing. A schedule can tell you when to check the account. It cannot protect the work unless someone owns the target and everyone who touches the funnel records what they did.

The old template never went back in a frame. The last thing I saw that Friday was the staging link loading with the form where it belonged. The cold cup was still on the kitchen table.

I still have the spreadsheet with its three tabs. The build tab is outdated. The checks tab is fine. Beside them now sits the change log I copy into every new launch, blank and waiting.