Monday morning, coffee, the weekly report. Cost per lead is up 40% on the week. You scroll until you find it: one campaign started spending twice its usual budget on Thursday afternoon and has brought in nothing since. Four days of spend, gone, and the report did its job perfectly. It told you exactly what happened. It just told you on Monday.
That is the problem with a weekly report. It answers “what happened last week,” which is a history question. The question that saves money is “what changed today,” and almost nobody has anything set up to ask it.
The platforms already built half of this
Google Ads has had automated rules for years. In Google's words, they “allow you to make automatic account changes based on chosen settings and conditions,” and “you can also use automated rules to trigger automated emails when specific conditions occur.” Meta Ads Manager has its own version.
Google also publishes a free script that does most of the thinking for you, the Account Anomaly Detector. It “alerts the advertiser whenever a Google Ads account is suddenly behaving too differently from what's historically observed.” The interesting part is how it decides what normal is: “stats for a Tuesday, 1 PM are compared with stats for 26 previous Tuesdays.”
That one design choice is the whole trick. Your account does not behave the same on a Tuesday as on a Saturday. Compare today to yesterday and every weekend looks like a crisis. Compare today to the same day in past weeks and only real changes stand out.
Where AI earns its place
Rules and scripts are good at one thing: noticing that a number crossed a line. They are bad at the part that actually eats your time, which is working out why. An email that says “cost per conversion above threshold” just sends you into the account to dig.
That digging is what an AI model is now good at. Give it yesterday's numbers by campaign and ad set, plus the same weekday from the last few weeks, and ask it one narrow question: what changed, where, and what is the most likely cause. What comes back is two lines instead of a dashboard:
The daily budget on Spring Retargeting was raised Thursday morning, and it has spent $612 since with no leads. Every other campaign is inside its normal range.
You still make the call. The model's job is to read the whole account every morning so that you only read the part that changed. If you can schedule a script and call an API, this is an afternoon of setup. If you can't, a no-code tool like Zapier or Make can pull the numbers into a sheet each morning and pass them to a model for the summary.
The alert most setups miss
Here is the failure that catches people who already have alerts. Most alerts watch cost per lead, and cost per lead is spend divided by leads. When leads drop to zero, there is nothing to divide by. Depending on the tool, the number goes blank, shows a dash, or quietly falls out of the check. The worst day your account can have, money going out and nothing coming in, can read as “no data” instead of an emergency.
The cause is usually not the ads at all. It is a form that broke, a tracking tag that stopped firing after a site update, or a thank-you page that moved. The ads keep spending, the conversions stop being counted, and the cost-per-lead alert has nothing to say.
So add one rule that does not depend on a ratio: if a campaign spent more than a set amount today and recorded zero conversions, tell me. It is the simplest alert on this list and the one most likely to save you a week.
Make it quiet enough to trust
An alert you learn to ignore is worse than none, because it trains you to ignore the real one. Google's script builds this in: “only a single email for an alert is sent per day.” A few more habits keep it quiet:
- Compare to the same weekday, never to yesterday.
- Set a minimum spend before anything can alert, so a $4 ad set doubling to $8 never wakes you up.
- Wait for the data. Google notes its statistics “can be up to 3 hours delayed,” so an early check is still looking at last night.
- One message per problem per day. If it fires again tomorrow, that is new information. If it fires every hour, it is noise.
What to set up this week
- Install Google's Account Anomaly Detector on your main account. It is free and runs from a template spreadsheet.
- Add the zero-conversion rule in Google Ads and in Meta: spend over your chosen amount, zero conversions, email me.
- After a week of alerts, add the AI summary, so each alert arrives with its likely cause already attached.
- Keep the weekly report, but read it for trends, not for problems. Problems should already have reached you.
Where we land on this
The weekly report is not wrong, it is late. Going from “read everything on Monday” to “hear about the one thing that changed, the day it changed” is mostly plumbing, and AI is what makes the message short enough to act on before your coffee is done.
The same gap exists in the traffic itself. Your alerts can tell you a campaign drove 400 visits and no leads. They cannot tell you who those 400 people were. That is the part we work on.