Sunday evening, first batch of Meta automated rules. Five of them. For each one the action dropdown asks the same thing: turn the ad off, or send a notification only? Every guide says "start with notifications". None says when to stop.

So all five stay on notify, and two weeks later you are still pausing ads by hand at 11pm.

Short answer: An automated rule should pause ads on its own only when a wrong action is cheap to undo, waiting for a human is expensive, and the metric it reads is thick with data when it fires. Spend running past a cap passes all three. A cost-per-purchase rule on a three-day-old ad passes none, so it only alerts.

The takeaways

Three questions sort every rule. How cheap is a wrong action to undo, what does an hour of waiting cost, and how much data sits behind the number at the moment the rule fires.
Spend and frequency rules can usually act; cost and return rules usually alert. Spend grows with every impression, while on four purchases a day, one sale fewer raises cost per purchase by a third.
Pausing is not a free undo. A paused ad set can lose delivery stability when it restarts, so "I can always turn it back on" is weaker protection than it sounds.

What decides whether a rule should pause or only alert?

Three questions decide whether an automated rule should pause or only alert, and I ask them in this order.

First, reversibility. What does a wrong action cost to put back? Lowering a budget is close to free. Ending an ad is close to permanent.

Second, the cost of waiting. A notification gets read some hours later. For a runaway budget at 2am, those hours are money. For a slowly fading creative, they cost almost nothing.

Third, how thick the data is at the moment of firing. Spend grows with every impression, so a crossed line means something happened. Cost per purchase on a small ad set rests on a few events, so a crossed line can be one quiet afternoon.

A rule earns the right to act when all three answers point the same way. Two out of three means alert.

Which automated rules should pause ads on their own?

The automated rules that should pause ads on their own guard against fast, expensive, unambiguous failures. Runaway spend is the clearest case: a budget typo or a scaling mistake burns money every hour you sleep, the spend figure is never thin, and a paused campaign is easy to restart in the morning.

Frequency is the second. Once an audience has seen the same ad three or four times, fatigue rarely reverses on its own, and frequency is a count of impressions, so it does not bounce around the way purchases do.

A landing page returning errors belongs here too, but ad-platform rules mostly cannot see it, so it needs monitoring outside the rule builder. I went through it in what to do when your website goes down mid-campaign.

Each of these catches something you would have paused the minute you saw it.

Which rules should only send a notification?

Rules that read cost per purchase, return on ad spend or click-through rate on young or small ad sets should only send a notification. Those metrics depend on a handful of events, so a single slow day crosses the threshold and the rule fires on weather.

My sanity check: an ad set spending €120 a day at three purchases sits at €40 per purchase. Two purchases the next day puts it at €60, half again above the target, and nothing about the ad changed. A kill rule set at €55 would have ended it.

The second group is anything that needs context the rule lacks. A drop in return on ad spend might be the creative, a tracking outage, a competitor's sale or conversions that have not arrived yet. You can check those in ten minutes. The rule cannot check them at all, so its job is to make you look.

Is pausing an ad a safe default for a rule?

Pausing an ad is safer than ending it, and less safe than most buyers assume. A pause keeps the ad and its history, so it is reversible in the narrow sense. Delivery is another matter: an ad set that stops and restarts can wobble while Meta's system finds its footing again, and that wobble is part of what a wrong pause costs.

So "it only pauses" is no reason to let a thin-data rule act. Ask what a wrong pause costs on this specific ad set. For a mature one with steady volume, usually a day or so of rougher delivery. For one still in learning, it can set the clock back. The details are in whether pausing a Facebook ad resets learning.

How do you move a rule from alert to action?

Move a rule from alert to action by counting how often you agreed with it. Leave it on notify and write down every firing and what you did about it. After two or three weeks you have a hit rate.

If you paused by hand nearly every time it fired, the rule already makes your decision and arming it only removes the delay. If you ignored half its firings, it reads noise, and arming it hands a coin flip your budget. Widen the threshold or lengthen the window, then count again.

One rule I keep regardless of hit rate: nothing kills an ad that has had too little time or spend to be judged. That one stays on notify forever.

Where software fits in this decision

I built the kill path in Adscalr around exactly this sorting, and the defaults reflect it. Rules start as recommendations, and full-auto is strictly opt-in. No rule may act on an ad younger than 5 days or with less than €200 spent. Ads at 1.5x return on ad spend or better can be paused but never killed, a kill can be undone for 30 minutes, and every rule evaluation is logged so you can count its hit rate yourself.

None of that decides for you which of your rules deserve to act. The automation layer gives you the dial and the record. Where you set it is still your call.

Should automated rules pause ads or just alert?