How to Use Birch to Pause Losing Ad Sets Automatically

The short answer: connect only the Meta accounts you mean to automate, build a pause rule at ad set level whose condition pairs a result threshold with a spend floor, schedule it every 15 minutes with alerts to Slack, put budget increases in a second rule of their own, let Birch AI draft further rules that a person checks and publishes, and read each rule's log after the first week.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

Every paid social team has someone who opens Ads Manager first thing to switch off what spent overnight without selling. That routine is a rule, and Birch, the renamed Revealbot, runs rules on Meta, Google, TikTok and Snapchat ad accounts as often as every 15 minutes. For what Birch includes and costs, see the Birch glossary entry.

Before you write the first rule:

1. Know what one check can cost you. For example, an ad set with a $40 daily budget, spending at an even pace, uses $40 x 15 / 1,440 = $0.4167 in the 15 minutes between two checks, so the schedule is rarely the leak. The leak is a floor set too high: under the pause example in Birch's pause rules article, "Website Purchases < 1 and Spend > $150 (Last 7 Days)", a non-converting ad set can spend $150 before the rule acts.
2. Start from the thresholds your team already uses. The target cost per purchase and the break-even ROAS someone checks by hand are the numbers to write into the rule.
3. Pick where you will watch it. A Slack channel for alerts and the rule's own log for every check it made.

Step 1: Connect only the accounts you will automate

Sign up without a card; the 14-day trial includes every Pro feature (getting started). Rules need Pro afterwards, from $99 a month on monthly billing for up to $10,000 of ad spend (pricing, checked on 26 October 2026). Go to Settings, Platforms, and connect Meta. An ad account counts as connected, and counts toward the bill, once it is assigned to a workspace. A rule engine billed on the spend it watches does better the more accounts you link, so link the ones you will automate and leave the rest out. Unlinking later turns rules into drafts and permanently deletes reports, custom metrics and launch templates for that account.

Step 2: Open the rule builder and set the scope

Select Meta in the platform selector at top left and click Create rule, then choose to build from scratch (creating rules). Give the rule a name your team will recognise, pick the ad account (up to five that share a currency), and set the level to ad sets. Under Filter, narrow it to what you mean, for example ad set status is active and campaign name contains "Prospecting". Filters do not support emojis, so match on text.

Step 3: Write the stop-loss condition with a floor

Add a Pause task and join the conditions with AND. Birch's examples pair the performance metric with a spend or impressions condition, because a metric with no data reads as zero by default (pause rules). Two forms work well:

• No sale after real spend: Website Purchases < 1 and Spend > $150, both over the last 7 days, as in Birch's pause examples.
• A bad day inside a good week: Spend > $15 today, ROAS < 1.5 today, and ROAS < 3 over the last 7 days, so one weak day alone does not switch off an ad set with a good week behind it.

Replace Birch's numbers with your own from the list you wrote.

Step 4: Schedule it, cap it and route the alerts

Under Schedule, choose Run every 15 minutes, or specific days and hours. Tasks that change something have a Frequency setting that caps how often one item can be touched, even if the rule checks more often. Send notifications to the marketing channel in Slack, keep the attribution window setting switched on as Birch recommends, and click Preview, which writes the rule out as a sentence you can read back (creating rules).

Step 5: Put scaling in a rule of its own

Birch's budget example raises the budget when ROAS > 3 today and spend has passed half the daily budget (increase budget rules). Build it as a second rule, not a second task in the first one. The help centre advises beginners to keep tasks in separate rules so each log stays readable, and it warns that budget tasks sharing one rule with a once-a-day cap block each other: only one of them can change an item's budget that day (once-a-day budget tasks).

Step 6: Let Birch AI draft rules, and publish them yourself

Click Birch AI at the top right and describe the goal, for example "generate a CPA protection rule for prospecting ad sets, cost per purchase above $40 over 3 days". Birch AI reads your workspace and creates the rule as a draft for you to review and edit before publishing (Birch AI). The person who owns the account checks three things before pressing publish: the threshold matches the written list, every metric condition has a spend or impressions floor, and the scope does not overlap a rule already live.

For a weekly audit from Claude, open Claude's connector settings, add a custom connector named Birch with the address https://mcp.bir.ch/mcp, and sign in (Birch MCP setup). Ask it to review the rules on your Meta account and flag overlaps, contradictory actions and thresholds with no volume guard, as Birch's MCP overview suggests. The MCP is in beta, returns what rules did rather than ROAS, and cannot create rules from Claude yet.

Step 7: Read the logs after a week

Hover over the rule's name and click the logs button. Each row is one run; green rows show items the rule affected, beside the count of items it checked. Hover a row to see the values Birch pulled from Ads Manager and which conditions matched, and use Search logs by item ID to find out why a given ad set was left alone (rule logs). Times show in your own timezone, which can differ from the account's.

What Birch's help centre warns about

• Conditions without a floor. With no data, metrics equal zero, so a CPA rule without a spend condition can act on noise.
• Budget tasks that cancel each other. Several once-a-day budget tasks in one rule produce one change per item per day.
• Overage on monthly plans. Spend above the plan limit is charged as a percentage, and upgrading does not remove overage already built up (additional charges). Through overage or the next tier, Birch's bill rises with your ad spend, whether or not its rules did any extra work that month.
• A partial rule list in the MCP. It currently lists only rules in the General category, without warning that others are missing.

Checking the rules against sales

A rule judges ad sets on the ROAS and cost per purchase that Meta reports, so it can switch off an ad set that was creating demand another channel was credited with. The free test is a holdout: leave one comparable campaign without rules for four weeks and compare cost per order from your shop system for both. If you later want that comparison running all the time, a causal attribution tool such as Causality Engine is built for it; disclosure: I founded it.

FAQ

How often can a Birch rule run?

From every 15 minutes up to every 72 hours, or on specific days and hours you set, according to creating rules.

Can Claude write Birch rules?

Not yet. Birch MCP manages existing rules and reads their logs; natural-language rule creation lives inside the Birch app, and Birch says any rule created from Claude in future will start as a disabled draft (Birch MCP setup).

Do the same rules work on Google Ads and TikTok?

Rules exist for Meta, Google, TikTok and Snapchat, each chosen in the platform selector (creating rules). Birch AI generates rules for Meta and TikTok.