A campaign doubles its spend on a Saturday morning. The dashboard shows it. In red, even. Nobody sees it until Monday.
That is the problem with dashboards. They wait. They hold the answer until someone shows up to ask the question.
Alerts flip that model. The system watches the numbers and speaks up the moment something breaks. The best alerts go one step further. They act.
Three tiers: wait, watch, fix
Think of monitoring as three tiers.
A dashboard answers a question when a person asks. An alert notifies a person when a number crosses a line. An auto-action fixes the problem itself, then tells the person what it did.
Most marketing teams stop at tier one. They build beautiful reporting, then staff it with hope. Someone will check it. Someone will notice. On a long weekend, nobody does.
The goal is to move every failure that costs money up the tiers. Watch it automatically. Fix it automatically where you safely can.
Thresholds: alert on decisions, not on movement
An alert earns its existence with one test. If it fires, will a specific person do a specific thing? If the answer is "we would just look into it," that is not an alert. That is a report.
Two kinds of thresholds do most of the work.
Absolute thresholds
Hard lines that should never be crossed. Daily spend above the cap. Zero conversions by noon in an account that converts every day. A landing page returning an error. An ad or a product feed disapproved.
These are easy to define and hard to argue with. Start here.
Relative thresholds
Comparisons against normal. Cost per lead running well above its trailing average. Monthly spend pacing ahead of budget. Click-through rate sliding against its baseline.
Relative thresholds need more care. Compare against the same day of the week, not against yesterday. A Sunday always looks broken next to a Thursday. Build the baseline into the alert so it does not cry wolf every weekend.
Ownership: one alert, one name
An alert sent to a group channel belongs to nobody. Five people see it, each assumes someone else has it, and the alert dies in plain sight.
Every alert needs exactly one owner. Not a team. A name.
Write each alert down in a simple register: what it checks, the threshold, who owns it, what the owner is expected to do, and how fast. That register is the difference between an alert system and a pile of notifications.
When the owner changes roles or goes on vacation, the register tells you what to reassign. Without it, alerts quietly go orphan, and nobody notices until one of them mattered.
Escalation: what happens when nobody answers
An alert without escalation is a suggestion.
The pattern is simple. The first notification goes to the owner. If nobody acknowledges it within the response window, it fires again to the owner plus a backup. If it is still sitting there, it goes to someone with the authority to act, a manager or the founder.
Match the escalation window to the cost of delay. Runaway spend escalates within hours. A slow drift in cost per lead can wait for tomorrow's review.
The point is not blame. The point is that a money-losing failure never depends on one person's phone being on.
Auto-actions: let the system pull the brake
Some failures should not wait for a human at all.
Pausing overspend
This is the clearest case. Set a hard daily cap above the planned budget. If a campaign blows through it, the system pauses the campaign, then notifies the owner with what it did and why.
A pause is reversible in one click. Overspend is not. When in doubt, build the brake.
Other good candidates
Pause ads pointing at a landing page that keeps failing a health check. Hold a campaign whose conversion tracking goes silent, because every optimization decision made on top of dead tracking is a bad one.
Where to draw the line
Automate what is reversible and safe. Pausing qualifies. Raising budgets, changing bids, and restructuring campaigns do not. Keep those in human hands.
We build AI into the middle of this layer. It drafts the diagnosis, summarizes what changed, and proposes the fix. A person approves anything that moves money. The machine pulls brakes. People hit the gas.
Alert fatigue is a design failure
If your team ignores alerts, the alerts are wrong, not the team.
The symptoms are familiar. A muted channel. A wall of unread notifications. The phrase "oh, it always does that."
The fixes are all design:
- Delete any alert that does not map to an action. Be ruthless about it.
- Tune thresholds until false alarms are rare. Every false alarm spends trust you will need for the real one.
- Split urgent from informational. Real-time pings are for money and breakage. Everything else goes into a daily digest.
- Review the register monthly. If an alert fired all month and nobody acted, fix the threshold or kill the alert.
A small set of trusted alerts beats a large set of ignored ones. Every time.
A starter set worth building
If you are starting from zero, build these first:
- A hard spend cap per account, with auto-pause.
- A zero-conversion check by midday, for accounts that convert daily.
- A tracking heartbeat that compares today's tag fires against a normal day.
- Uptime checks on your highest-spend landing pages.
- Disapproval alerts for ads and product feeds.
- A pacing alert against the planned monthly budget.
Each one guards against a failure that costs money silently. Together they cover most of the expensive surprises we see inside ad accounts.
This is the operational layer underneath your marketing, the same layer we build in our marketing systems work. Alerts sit next to reporting that writes itself and the account structures that make thresholds easy to set in the first place. When we manage paid media, this layer is part of how we run it, documented in the open, owned by you.
Find out what your alerts would have caught
Look back at your last few expensive surprises. The overspend nobody caught in time. The tracking that died quietly between reports. Almost every one of them could have been caught by an alert that takes an afternoon to build.
Our Performance Audit maps how your marketing actually runs, including the failures that happen silently between reports. You leave with a dollar-weighted plan you keep, whether or not we build anything together.