Routines: AdsGuru's automated rules

Updated November 12, 2026

What a routine is

A routine is a recurring optimization job that ships with its decision logic already built: when to look, what to compare, what counts as enough evidence, and what change to propose. You tune its thresholds; you do not write its logic.

Each routine produces proposals, and approval mode is the default: a routine's output lands in your queue with the change, the reason and the evidence, and applies only when you approve it. Once a routine has repeatedly earned approvals, you can promote that one routine to autopilot with caps, and demote it any time. Every applied change is logged either way.

The routines cover the recurring work of ad account maintenance: pause underperformers, clean search terms, add negative keywords, pace and reallocate budgets, and rotate fatigued creative.

What classic automated rules are

Classic automated rules are if-then statements you write in a platform or a third-party tool: if CPA exceeds $40 for 3 days, pause the ad; if ROAS exceeds 3 for a week, raise budget 15%. Google Ads and Meta both have built-in rule engines, and tools like Birch build businesses on more powerful versions.

Rules are genuinely useful, and the format has real strengths: perfectly predictable, instant, and exactly as smart as the person who wrote them on the day they wrote them. That last property is also the weakness. The rule executes its letter, not its intent, in every situation including the ones you never thought about.

The other cost is maintenance. A useful rule set grows: exceptions for the learning phase, exceptions for sale periods, different thresholds per campaign type. Someone has to own that library, test it, and update it when the account changes. In practice the rule library becomes a small codebase.

Why if-then rules misfire

Rules misfire because ad accounts are contextual and rules are not. The common failure patterns:

Small samples. 'Pause if CPA above $40' fires on an ad with 3 conversions of noisy data, killing an ad that was fine. Statistical judgment, how much evidence is enough, is exactly what a bare threshold lacks.

Learning phase ignorance. Meta ad sets perform erratically during learning; rules that act on learning-phase data pause or boost things based on numbers that were about to change anyway, and each edit can reset learning again, compounding the damage.

Sibling blindness. Pausing the worst ad in an ad set shifts its spend to siblings. A rule judges each ad in isolation; whether the siblings are better is the question that actually matters.

Threshold cliffs. CPA of $39.99 is fine and $40.01 pauses, applied with equal confidence to a campaign spending $10/day and one spending $1,000/day.

Stacked rules fighting. A pause rule and a budget rule interacting can oscillate: budget rises, CPA ticks up, pause fires, budget shifts, repeat. Nobody designed the oscillation; it emerged.

How routines build the judgment in

Routines run the same checks a careful human would run before acting, automatically, every time:

• Evidence thresholds: enough spend and conversions before judgment, scaled to the account's volume rather than a fixed count.
• Trend significance: movement is compared against the account's own normal volatility, so one bad day is noise and six are signal.
• Learning awareness: proposals avoid acting on learning-phase data and size changes to avoid unnecessary learning resets.
• Sibling context: a pause proposal checks where the spend will go and whether that destination is actually better.
• Caps: budget movements are bounded per day, with floors and ceilings per campaign that you control.
• A reason, always: every proposal states its cause and evidence, so you can verify instead of trust.

The routines AdsGuru ships

Every routine starts in approval mode. There is no master switch that hands the account over; trust is granted per routine, and structural changes never run on autopilot at all.

Tuning a routine

Each routine exposes its thresholds, so you set the risk appetite without writing logic: the CPA or ROAS trigger levels, minimum spend before an ad is judged, the evidence window in days, maximum budget movement per day, and per-campaign floors and ceilings.

Tune by outcome. If a routine proposes too much, tighten its evidence requirements or widen its target tolerance; if it is too quiet while you can see waste, loosen them. The proposals themselves are the feedback loop: each one shows which threshold it tripped, so you learn exactly which knob to turn.

Defaults are deliberately conservative, sized so that a brand-new user who approves everything would still be making defensible changes. Most users tighten toward their taste over the first month.

Threshold changes take effect on the next run, and the change itself is recorded, so a quarter later you can answer why February's proposals looked different from January's. Tuning without a record is how teams end up with settings nobody can explain.

Approval, autopilot and caps

The lifecycle of trust runs one way, deliberately. Every routine starts in approval mode: proposals queue, you decide. When a routine has produced, say, three weeks of proposals you approved without edits, promoting it to autopilot saves you the clicks without changing what would have happened.

Autopilot always runs inside caps: maximum daily budget movement, floors and ceilings, minimum evidence. A change that would exceed a cap comes back as a proposal instead of being applied. Autopilot changes are logged and reported identically to approved ones, so the daily note always shows what ran, and demotion back to approval mode is one click.

This is the honest trade of the whole product: you give up writing your own automation logic and get judgment that arrives pre-built, gated behind your approvals, with its reasoning shown. If that trade is wrong for you, the next section is the honest alternative.

When a rule builder is the better buy

Choose a rule builder when your edge is the logic itself. Teams with genuinely custom playbooks, unusual conditions tied to external data, inventory levels, weather, flight schedules, or a full-time buyer who wants to encode their own decisions precisely are rule-builder customers, and tools like Birch serve them well. Checked November 2026, Birch Pro starts at $99/month up to $10,000 of monthly ad spend and rises by spend bracket; RevealAds-style rule tooling generally prices by spend as well.

Choose routines when the maintenance work is standard and the scarce resource is time and attention: most stores and lead-gen businesses under a few hundred thousand of monthly spend. You can also run both: AdsGuru does not require exclusivity, though two systems editing one account should have clearly divided jobs to avoid fighting each other.

You can tune every AdsGuru routine's thresholds, but free-form rule authoring is deliberately not the product. That focus is what keeps the reasons readable and the behavior predictable.

Common questions

Related