Run Google and Meta ads on a fixed rhythm

The account in this guide: the same Australian education provider as in my Google Ads audit guide and my Meta lead ads guide. I do this work inside someone else's business, on a monthly cycle with a short brief every morning. The cycle has run longest on the Google side, so most of the examples come from there. I can't name it, but the numbers are real.

01 Watch every week, change almost nothing

The weekly check catches what's broken or drifting. On this account a lead takes weeks to become a sale, so one week can't tell you whether a campaign works. It can tell you whether tracking still fires and whether spend is going where you planned.

Drift is the one people miss. In one three-week stretch on this account, spend on the course with the best search returns in the last audit fell by about two fifths. Nobody decided that. In the same stretch a weaker campaign nearly doubled its spend, and the audit flagged that as drift too. It wasn't drift. I had raised its budget the month before, and the audit hadn't read the log. A weekly look at pacing by campaign, with last month's changes beside it, would have caught both in week one.

Judge pacing over the week, not the day. Google can spend up to twice a campaign's average daily budget on a single day, and up to 30.4 times it over a month, so one heavy Tuesday means nothing.

Two kinds of change are allowed mid-month. Repairs: broken tracking, disapproved ads, a dead landing page. And safety: runaway spend, or a junk search term with no sales against it. Everything else goes on the list for the monthly plan.

An AI writes this account's morning brief from published tables. It isn't allowed to calculate anything. Every figure has to appear in the tables, or the brief says it's missing. One recent brief flagged two courses with leads and no enrolments, then added that this was not grounds for a budget cut.

02 Audit once a month, starting with last month's log

The monthly audit uses the method in my Google Ads audit guide, and for Meta lead ads, the one in my guide to measuring lead ads against sales. Two things turn it from a one-off into a rhythm.

First, it reads last month's change log before anything else. Each change went in with an expected effect and a date to judge it, so the audit starts by asking whether those changes did what they were meant to. On this account the audit didn't do this at first, and in July it flagged a budget increase I'd made on purpose as drift. Reading the log first is the fix.

Second, it has a freshness gate. If the sales data is more than a week behind the end of the window, the audit goes out marked provisional and budget moves wait. The Google Ads audit guide shows what that looked like here.

Before scoring, check that spend matches the platform totals. Then score and flag each campaign the way FM-01 does.

03 Plan every change with its reason

The plan turns the audit into a short list. Each line carries five things: the change, the evidence for it, the step size, the effect you expect, and the date you'll judge it. Before the plan is final, its spend figures have to reconcile with the audit. If they don't tie out, the plan stops there.

The reason matters most later on. Google's change history keeps two years of changes, with who made them and what changed. It doesn't keep why. Three months on, "budget down 35%" tells you nothing. Next to its reason and expected effect, it tells you whether the change worked and when to undo it.

Three rules keep the plan honest:

1. Move budget in steps of about 20%. Bigger jumps make it hard to tell what caused what, and as the file below shows, they don't always buy more sales.
2. One lever per campaign per cycle. Change the budget or the bid target or the keywords. Change all three and you won't know which one worked.
3. Don't touch a campaign that's learning. Google puts a bid strategy into learning when it's new or reactivated, when campaigns, ad groups or keywords are added to it or taken out, or when the conversion goals it bids towards change significantly. Budget and target changes don't trigger it. Google says the status lasts up to seven days, that performance can take one to two conversion cycles to settle, and that you should judge results over two conversion cycles or two weeks, whichever is longer. A conversion cycle is the time from click to conversion. If you bid on sales that take weeks to close, the wait is weeks.

Then check the plan against what went wrong before. This account's plan stage reads its own past lessons first, so a move that overspent in one state last time doesn't get proposed again.

04 Approve, then apply

Nothing the AI writes touches the account. The plan is a draft. I read each line, approve it or strike it, and apply the approved changes in the platform myself. If the approver can't say why a change is there without opening the audit, the reason isn't written well enough.

Before anything goes live, record two things: the value you're changing from, and what would make you reverse it. When the two weak campaigns were paused, the log set a review 60 days out and said what would reopen the decision: if total sales for that course, paid and organic together, fell more than 10% against the previous 60 days, the paused spend had been finding buyers nothing else reached. A rollback trigger written before the change is a test. One written afterwards is an excuse.

For bigger changes, such as a new bid strategy, use an experiment. Google's custom experiments split a campaign's traffic and budget between the original and a trial version, and if the trial wins you apply its changes to the original. Ending the experiment is the rollback. They're available for Search, Display, Video and Hotel campaigns, not Shopping or App.

Watch Google's auto-applied recommendations too. They're opt-in, but once on they apply regularly: changes nobody approved. Turn them off, or read their History tab every week.

05 Check it happened, then judge it on the date

Check that each change actually took effect. In the same June round, the budget on the account's Performance Max campaign was raised. A month later it was spending about half of its new budget. The plan had assumed the extra budget would be spent. The July check found it wasn't, and why went on the list for the next review. A week after any budget change, compare actual spend with the new budget.

Then wait for the recheck date. On it, each change gets one call: keep, extend by another step, or roll back to the value in the log. Rollback is cheap when the before value is written down.

The same rhythm on Meta

The cycle is the same. Learning works differently. Meta says an ad set leaves its learning phase after about 50 results in the week after its last significant edit, and recommends waiting until then to edit. It also says a large budget change can count as a significant edit and restart learning. Ads Manager has a "Last significant edit" column; add it to your view and read it before you touch anything. A lead ad set optimising towards a sale stage may never reach 50 a week, and Meta flags an ad set that won't get there as Learning limited. Change those rarely. Every week, count the leads still waiting for a first call. On this account, the queue was the first thing to rule out when Meta numbers moved.

Where AI fits in the workflow

On this account each stage of the monthly cycle has a written contract: what it reads, what it writes, and when it stops. A script pulls and reconciles the numbers. A model reads them, drafts the audit, groups search terms into themes, drafts the plan with a reason on every line, and writes up the change log entries. A stage that finds totals that don't tie out stops and says so. Stale sales data marks the audit provisional and holds every budget move.

I decide everything that spends or removes money: every budget move, every pause, every negative keyword, whether a reason still holds on its recheck date. The model can read the account data but not change it, and it doesn't know that a campaign which looks weak is feeding the search terms another one lives on.