Rules

A rule has three parts:

• A condition: a metric (spend, CPA, CPC, CTR, ROAS, or purchases) compared against a threshold over a window of 1 to 30 days. For example, CPA above 30 over the last 7 days.
• Guard metrics: minimum spend and minimum purchases the entity must have in the window before the rule may fire. Guards stop rules from reacting to statistical noise, like one bad day on two clicks of data.
• An action: what to do when the condition holds.

• Notify: send a notification, change nothing.
• Pause the campaign, ad set, or ad.
• Activate it again.
• Cut budget by a step, never below a floor you set.
• Raise budget by a step, never above a ceiling you set.

The floor and ceiling are hard limits: no sequence of rule runs can push a budget past them.

Every rule starts in manual approval mode. When it fires, the proposed action goes to the Pending actions queue together with an AI-written explanation of why the rule fired and what the numbers looked like. You review each proposal and click Apply or Dismiss; only Apply touches the ad platform.

A rule you have come to trust can be switched to auto mode, where its actions apply immediately and you are notified after the fact. This is opt-in, per rule.

Cooldowns keep rules from thrashing: after a rule acts on an entity, it waits out its cooldown period before it may act on that entity again.

Before trusting a rule, run it against history. A dry run replays the rule over your last 14 days of data, read-only, and shows when it would have fired and what it would have done. No notifications are sent and nothing is changed.

• Natural-language builder: describe the rule in plain words, like pause any ad set that spends more than 50 with no purchases over 3 days, and the AI drafts the structured rule for you to review and save.
• Suggested guardrails: pimads looks at your last 28 days of performance and suggests sensible starter rules with thresholds fitted to your actual numbers.

Both are drafting aids: what the AI produces is only a proposal until you save it, and even then it obeys the approval queue like any hand-written rule. More on the models and data on the AI features page.

Separate from your own rules, a daily watchdog compares each campaign against its own trailing-week baseline and flags:

• spend spikes,
• delivery drops,
• CTR collapse,
• CPA drift.

Each finding comes with a short AI note explaining what moved and by how much. The watchdog only notifies; it never changes anything.

Rules are evaluated hourly on the server. You can also trigger an immediate evaluation with Run now on the Rules page, useful right after creating or editing a rule.