Tools That Automate Creative Testing and Relaunch the Winners

Software that automates creative testing loops and feeds winners back into ad platforms automatically comes in two kinds, and the difference decides what you are buying. Rule engines such as Birch and Madgicx hold write access to the ad account, so their rules can pause ads, change budgets and, in Birch's case, duplicate winners unattended. Creative analytics tools such as Motion read the account and report what won. The ad platforms sit between the two: Meta, TikTok and Google each document a split test, and in most of the pages we read the step that applies the result belongs to a person.

In the vendor documentation we read on October 7, 2026, we found one split test that can apply its own result with nobody in the chair, Google's custom experiment, a campaign-level test covered below. In every other product we read about, you assemble the loop from a test method and a write path, and the test method you trust will shape your return on ad spend more than the vendor's name will. We ran no hands-on trials; everything below comes from what the vendors publish.

What a closed creative testing loop has to do

To automate creative testing loops and feed winners back into ad platforms, four jobs have to run unattended:

1. Launch. New creative enters a test structure with a fixed budget, naming scheme and audience.
2. A winner rule. Something decides, by a stated method, which ad won.
3. Promotion. The winner is copied into the campaigns that carry the budget.
4. Retirement. Losers are paused, and winners that fade later are paused too.

Launch, promotion and retirement are plumbing: API calls that create, duplicate and pause. The winner rule is a judgment, and it is where products differ most. A threshold (say, a cost per purchase under $32 after eight purchases) is cheap to automate and easy to fool, because nothing guarantees each ad the same audience or the same spend. A split test divides the audience first, which costs time and budget and buys a cleaner answer.

What Meta, TikTok and Google already automate

Creative testing automation inside the ad platforms, as Meta, TikTok and Google document it in 2026, covers the test and the pause-or-budget rules, while feeding winners back is still a person's job in most of the pages we read.

TikTok states its winner method most plainly. TikTok's split testing page (last updated January 2026) says the audience is split into two equal groups and that the feature "is scientifically designed to perform an accurate A/B test with a 90% confidence rate to determine which ad group performs better." Promotion is manual and short: "At the end of a split test, you can simply choose to continue running the winning campaign or ad group, optimizing towards your goals with just one click." Retirement can run unattended, since the examples TikTok gives for automated rules include "Pause a low-performing ad when it hits a particular metric" and "Increase your budget based on return on ad spend (ROAS)".

Google's Performance Max version works at the asset level and is still labeled Beta. Google describes Performance Max A/B testing of assets as a comparison of two asset sets inside one asset group, with a traffic split you define and an end date set by its Experiment Guidance System; the page recommends running "for a minimum of 4 to 6 weeks", and the assets in the asset group under test cannot be edited, added or removed until the experiment ends. The apply step is manual: "When you select Apply experiment, it updates your original asset group with the assets from the treatment group." Unchecking the default "Keep control assets in campaign" option removes the A set, so one click by a person both promotes and retires. Google's custom experiments, which test changes to Search, Display, Demand Gen and Video campaigns, are the exception we found: Google says that when the experiment beats the original, "you can apply your experiment to the original campaign (manually or automatically)." The same page sends Search ad creative tests to ad variations and video creative tests to video experiments. Separately, Google says of automated rules: "With rules, you can change your ad status, budget, bids, and more."

Meta documents both halves for developers. Meta's split testing guide says the Marketing API divides the audience into groups that do not overlap, advises testing one variable at a time, and tells you to pick the variant with the highest efficiency metric for the objective (for a conversions objective, the ad set with the lowest cost per action). We found no step in that guide that applies the winner for you. The write path is the Ad Rules Engine, which Meta describes as offering schedule-based rules and trigger-based rules, the second kind evaluated as soon as the relevant ad data changes. For scheduled rules, Meta lists daily, hourly, half-hourly and custom schedules, and its example rules pause matching objects or raise their budget by 10%.

Which third-party tools change the account and which only report

AI creative testing automation tools for Facebook, TikTok and Google Ads divide on one question: can the product change the account, or does it only read it?

Birch documents "automation rules for 4 different platforms - Meta Ads, Google Ads, Snapchat Ads and TikTok Ads", with actions listed as "pause, start, increase/decrease budget or bid, duplicate, etc." and a schedule where "the options vary from 15 minutes to 72 hours." The duplicate action is the nearest thing to a promotion step, and Birch documents its limits. Birch's FAQ, dated May 30, 2024, says: "At the moment duplication for Meta Ads works within a parent item only, so you can duplicate only within the same ad set or campaign." Birch's TikTok guide says "Duplication is currently available at the campaign level". We found no Birch rule action that copies one winning ad into a separate scaling campaign, so confirm what duplication can do on your platform during a trial. On Birch's pricing page, the Pro plan (the card that lists automated rules) is $99 a month on monthly billing for up to $10K in monthly ad spend, and $249 for up to $30K.

Madgicx aims its optimization product at Meta ("Optimize Your Meta Ads with AI"), with bulk budget increases and decreases and "automated rules created by the pros"; it says its rules trigger in real time and puts the market standard for recurring checks at 15 to 30 minutes. Madgicx sells one plan priced by monthly ad spend. On monthly billing it is $45 a month under $1K of spend, $129 at $5K to $10K and $329 at $20K to $30K; above $30K the page reads "See price inside the app". Its One-Click Report also monitors Google Ads and TikTok; we found its automation described for Meta only.

Motion also prices by spend: Starter is $750 a month for up to $50K in monthly ad spend, and Pro is $1,200 a month above that. Both list analysis work, such as "Routines: daily & weekly reports, iteration finders, alerts", and we found no launch, pause or budget action on either. The custom-priced Growth plan, for brands spending over $125K a month, is the one that lists "Closed-loop pipelines built with you, signal to launched ad".

The AI layer on both rule engines drafts or recommends, and a person applies the change. Birch says you can ask Birch AI to "Build a creative testing workflow", and the generated rules "are created as drafts, allowing you to review and edit them before publishing." Madgicx's AI Marketer audits the account daily and offers changes you apply in one click.

At $10K of monthly spend, our arithmetic from those pages puts the rule engines at about 1% of spend (Birch Pro, $99) and about 1.3% (Madgicx, $129), and Motion Starter at 7.5% ($750). Count the fee as acquisition cost when you compare setups on CAC.

A winner pipeline written as three rules

Automated creative testing software for Facebook and TikTok ads becomes a winner pipeline once three rules run against a test campaign: one kills, one flags winners for promotion, one retires. The example below is hypothetical, and every number in it is an assumption.

Assume an $80 average order value, a 2.5 target ROAS and a 2.0 break-even ROAS. Target cost per purchase is $80 / 2.5 = $32, and break-even is $80 / 2.0 = $40. Our ROAS calculator is a further resource for working those out from your own margins.

How much evidence the promote rule has

When you automate a creative testing loop and send winning ads to the ad platforms' scaling campaigns, the promote rule acts on thin evidence: in the example it fires on as few as eight purchases, or $256 of spend at the target. Treat it as a screen. A creative that drew the cheapest slice of a shared audience can clear the bar and then disappoint at five times the budget. For a decision that moves serious budget, confirm the screened winner in a platform split test before the scaling rule adds spend, and raise budgets in steps, as covered in how to scale paid ads without killing ROAS.

Every rule above reads platform-attributed purchases. A winner by attributed cost per purchase can still be collecting sales that would have happened anyway, which is a question for incrementality testing and a separate one from which ad won.

When the platforms' own features are enough

The native features cover the loop when you run one platform, add a handful of new ads a week and someone opens the account most days. A split test settles the decisions that matter, native rules handle the kills and the retirements, and promotion is a short manual step.

A rule engine earns its fee in three situations. The first is duplication that has to happen unattended: in the platform documentation we read, the rule examples pause ads and change budgets or bids, and we found no native rule action that duplicates a winner. Birch's duplicate action does, inside the same ad set or campaign on Meta. The second is one rule set that has to span Meta, Google, TikTok and Snapchat. The third is launch volume that has outgrown manual setup, the gap that launch features such as Birch's Stage ("Create ads from Drive media by duplicating your setups") are sold to fill.

An analytics tool answers a different question, which is why winners win, and the answer feeds the launch step with better briefs. Buy it for that reason, and budget for a person to press the buttons.

Write the winner rule down before you compare vendors: the metric, the minimum volume, the window, and who or what applies the result. If you would like a second pair of eyes on that rule, our paid media team is a further resource.

Sources

• Meta: Split Testing (Marketing API guide)
• Meta: Ad Rules Engine
• Meta: Schedule Based Rules
• TikTok: About Split Testing (last updated January 2026)
• TikTok: About Automated Rules (last updated February 2026)
• Google: About Performance Max optimization experiments: A/B testing assets (Beta)
• Google: Set up automated rules
• Google: About custom experiments
• Birch: Creating automated rules
• Birch: Pricing
• Birch: Can I duplicate ads into a new ad set? (FAQ, May 30, 2024)
• Birch: How to automatically duplicate TikTok campaigns
• Madgicx: Optimization
• Madgicx: Pricing
• Motion: Pricing