Reporting Answers the Wrong Question
A small business usually monitors its ads the way it monitors a bank balance: it looks at the total and decides the total is fine. Totals are almost impossible to read that way, because two opposite things cancel out. A paused campaign pulls the number down, and an edited daily budget pushes it up, and the report shows the arithmetic rather than either event.
The platform keeps a record of exactly those events. Changes made through the interface or the API are held for the past thirty days, and each entry carries the resource that changed, the type of resource, what operation was applied, and when. That record is the difference between knowing the account drifted and knowing what moved, and it is what a monitor should read first.
There is a second, quieter reason. Most small businesses do not have one person with full ownership of the account. The owner set it up, the office manager edits it when something looks off, an agency touches it monthly, and a technician's spouse who helps with the books has the login. The monitor is the shared memory that none of them have.
The reframe
The question is not whether the account is spending correctly today. It is whether anything moved since the last time you looked.
What Goes Undetected, and What It Costs
Move the sliders to match your account. The model shows how many changes happen in a month, how many go unnoticed, what the unnoticed ones are worth across a year, and the office hours recovered by reading daily instead of checking manually.
Changes per month
39
across 6 active campaigns
Nobody notices
18
45% of all changes
Annual exposure to unnoticed changes
$38,880
$3,240 per month at $180 each
Office hours recovered
12.8
per month, manual checking versus monitored
Detection quality
55%
of changes the monitor would name
What actually changes in an account
A pause is the one drift that stops spend entirely. It is either a deliberate change somebody forgot to tell you about, or an accident, and both need to be known today.
The most common drift. A daily budget edited during a busy week can quietly double the month. It is invisible in a weekly report that only shows totals.
The ad keeps its row in the report and stops earning. Spend on that group falls, so the totals look like a slow week rather than a broken ad.
A new campaign, a new ad, a new keyword. Often good news. Always worth knowing, because the person who added it and the person paying for it are not always the same person.
Geography, audience, match type, negatives. The slowest drift to notice and the most expensive to leave, because it changes who sees the ad rather than how much it costs.
Most days, nothing happens. A monitor that cannot stay quiet on these days is a monitor nobody reads by the second week.
The six buckets sum to all changes by construction, and 86% of them are real drifts. The last bar is the one that decides whether the monitor survives: a monitor that cannot stay quiet on the days nothing changed is muted by the second week, and then it is worth nothing at all.
Illustrative estimates from the inputs above. Your own numbers will differ; these are not measured results or client data.
The Monitor, Stage by Stage
Six stages from reading the change record to handing the finding to the outcome loop. Each names the system it touches, the payload it leaves behind, and where a naive monitor breaks.
Target System
Google Ads API reporting and change surfaces
State Read From the Platform
The monitor reads two things: the numbers, and the record of what changed. The change record is the more useful of the two, because it carries the resource name, the type of resource, the operation, and the time, which together are enough to name the change rather than merely detect its effect.
The monitor reads only the metrics, so a paused campaign and a bad week look identical until the month is over.
Stage payload
Reading the Change Record, and Reading the Numbers
The change record is a queryable resource, not a log file. Each row names the resource that changed by its resource name, the type of that resource, the operation applied, and the time of the change. It also carries the old and new versions of the resource, which is what lets a monitor say a daily budget moved from one value to another rather than merely that something happened.
The metrics come from a separate read, and the unit matters: cost is reported in micros, where one unit of currency is one million micros. A monitor that formats micros as dollars produces an account that looks ten thousand times more expensive than it is, which is the kind of mistake that gets an alert channel muted in a single morning.
The two reads are only useful together. Metrics alone cannot distinguish a deliberate pause from a collapse in demand. Change records alone cannot say whether the change cost anything. A monitor that reads both can say a specific campaign was paused on a specific date and that the pause is running at a specific monthly rate.
The read in one line
What the Monitor Does Alone, With You, and Never
Reading is free. Acting is not. The monitor can watch everything and change nothing, and that boundary is what makes it safe to point at a live account.
Autonomous
- Read campaign, ad group, ad, and keyword change records
- Read cost, click, and conversion metrics for the same window
- Compare today's state against the last stored state
- Classify a drift by cost direction and reversibility
- Assemble the alert with the old value, the new value, and the monthly impact
- Close an alert a person has acknowledged
Supervised
- Propose a budget correction back to the previous value
- Propose re-enabling a campaign paused by accident
- Propose a pause on a campaign whose cost per booked job has drifted
- Propose the review cadence for each drift tier
Never
- Change a budget, a bid, or a status on its own
- Pause or enable a campaign without a person deciding
- Rewrite an ad, a headline, or a landing page
- Assert why a change was made when the record does not say
- Import a conversion value no financial record supports
- Alter the account's conversion actions
What This Is Not
Not an autopilot for the ad account
Monitoring and managing are different jobs. This reads the account and tells you what moved. It does not decide what the budget should be, which ad is better, or whether the offer works. Those remain a person's call, and should.
Not a substitute for knowing your cost per booked job
A monitor can tell you a budget moved. Only the outcome loop can tell you whether the spend bought work. Watching the spend without connecting it to the jobs is watching the smaller half of the problem.
Not a way to alert on everything
A monitor that reports every change is a monitor that gets muted. The value is in the ranking, and the ranking means most changes are recorded silently and only the ones with money attached reach a person.
How to Sequence the Build
The order matters, because a monitor that alerts before it can stay quiet will be switched off before it catches anything.
Week 1
Store the state
Record the account's structure and settings daily. Nothing alerts yet. This is the baseline every future comparison needs, and it cannot be backfilled.
Weeks 2 to 3
Read the change record
Pull the platform's own record of changes and diff it against the stored state. Get the micros conversion right before anything reaches a person.
Weeks 4 to 5
Rank, then alert
Classify each drift by cost and reversibility, and route only the top tier to a person. Most changes should be recorded and never sent.
Ongoing
Connect it to outcomes
Hand the spend-side finding to the loop that imports booked outcomes, so the account is watched from both directions rather than one.
Questions About AI Ad Monitoring
Where This Fits in Marketing and Reputation
This page is the spend-side read. It defers the write direction to the offline conversion loop and the cluster framing to marketing automation.
Find Out What Changed in Your Ad Account Last Month
Give us read access to one account. In thirty minutes we will show you the changes made in the last thirty days, which of them were never noticed, and what the unnoticed ones are running at. You keep the analysis whether or not you hire us.
