How Bach.ai Auto-Tags Your Meta Creatives for Performance Pattern Mining

You have 84 ads running across 12 ad sets. Some are winning, some are losing. The winners share patterns — same hook type, similar visual framing, comparable CTA style — but you have no easy way to see those patterns because nobody has the time to tag every creative by hand.

Manual creative tagging is a job most performance marketers skip. It’s tedious, the taxonomy drifts across team members, and by the time you’ve tagged 80 ads, the first 30 are already stale.

Bach.ai tags your image ads for you, on 8 dimensions, and then looks for the tag combinations that beat or trail your account average. Here is what it tags, when it runs, and what it does not do.

Why Creative Pattern Mining Is Invisible Without Tags

Without tags, your creative library is just a list of ads. With tags, it becomes a structured dataset you can query — ‘which hook type wins on cold audiences’, ‘which CTA style beats the others on retargeting’, ‘which visual subject correlates with highest ROAS in Q4’.

The reason most brands never get this clarity:

• Manual tagging doesn’t scale. Tagging 1 ad takes 90 seconds if you’re being thorough — hook type, format, copy angle, CTA, visual, social proof, urgency. For a brand with 80 active ads, that’s 2 hours of work per refresh cycle, every week.
• Taxonomy drifts across people. One marketer tags ‘pain-point hook’, another tags ‘problem-led hook’, a third tags ‘PAIN’. Three weeks in, your dataset is unusable for cross-comparison.
• Patterns require volume. You can’t see ‘hook type X wins’ from 5 ads. You need 30+ ads tagged the same way. Most brands never accumulate that volume cleanly.
• The insight window closes fast. Patterns that win in March may not win in June. By the time you’ve tagged enough to spot the pattern, the pattern has shifted.

This is why most brands’ creative strategy is gut-feel iteration. ‘This concept worked, let’s try variations.’ That’s better than nothing, but it’s not pattern mining — it’s pattern guessing.

How Bach.ai Auto-Tags Creatives

Bach.ai sends each ad’s image (for a video ad, its thumbnail) to a vision model with a fixed taxonomy and stores one value per dimension. It reads the picture only: it does not read the ad copy, captions or video frames.

The 8 dimensions

1. Asset type: image, video, carousel, collection.
2. Visual format: UGC, studio, lifestyle, flat lay, product shot, before-after, testimonial.
3. Hook style: question, statistic, testimonial, before-after, bold claim, curiosity gap, social proof.
4. Creator persona: founder, influencer, customer, professional, AI avatar, none.
5. Messaging angle: discount, quality, social proof, scarcity, convenience, emotion, comparison.
6. Offer type: percentage off, buy-one-get-one, free shipping, bundle, limited time, no offer.
7. Product focus: single product, collection, lifestyle, brand story, review highlight.
8. CTA style: shop now, learn more, get offer, sign up, custom, none.

Most dimensions also allow “other” when nothing fits. Because one fixed taxonomy labels every ad, the tags stay consistent across the account in a way a team tagging by hand rarely manages.

When tagging runs

Tagging runs in the background after a Brand Discovery run on your account (a paid-plan feature), and you can start it yourself on the Creative Intelligence page, in the Analysis tab’s Tagging view, with Run Auto-Tagging, or Re-tag All once ads are tagged. It does not run on every data sync, so an ad you launch today is tagged the next time tagging runs.

What You Actually See in the Product

Tagged creatives

Each tagged ad shows its eight tags, and you can filter the set by any dimension, for example every UGC-format ad with a discount angle.

Winning patterns

Bach.ai groups ads by the combination of visual format, hook style, messaging angle and CTA style, over the last 90 days. A combination needs at least 3 ads to count. If its ROAS is more than 30% above your account average it is marked winning, and the Creative Intelligence page shows your top winning combinations. Behind the scenes it also marks combinations more than 30% below average as losing, which feeds how it scores a creative, and an above-average combination whose ROAS in the second half of the window was more than 25% below the first half as fatiguing. Combinations close to average are left out, because they tell you nothing.

What It Is Worth

Tagging by itself doesn’t save money; acting on the patterns might. What they can change:

• Better-informed creative briefs. A brief built on a combination your own account has already rewarded is a better bet than one built on instinct.
• Sharper reads on losers. When a new ad underperforms, its tags show which combination it belongs to, so you can check whether that combination is among your winners.
• Spend that follows the evidence. Shifting production toward winning combinations is the gain, and it is account-specific, so measure it against your own baseline rather than an expected lift.

The creative testing framework explains how to act on these patterns across a four-variant test loop.

Setup: What You Need to Do

1. Sign up at app.wittelsbach.ai.
2. Click Connect Meta and complete the sign-in with Facebook.
3. Open the Creative Intelligence page, go to the Analysis tab, open Tagging and press Run Auto-Tagging if your ads are not tagged yet.
4. Review which combinations are winning, and brief your next round on them.

Tagging and patterns are read-only. Nothing changes in your ad account unless you approve a specific change.

Frequently Asked Questions

How accurate is the auto-tagging?

It is most reliable on what is visible in the image, such as asset type, visual format and product focus, and least reliable on judgement calls like messaging angle, where two people may disagree too. Treat the tags as a model’s reading of one image and spot-check a few before you brief on them.

Can I add my own tags on top of the auto-tags?

Not today. The eight dimensions and their values are fixed, which is what keeps them comparable across ads. If you need brand-specific labels, such as a collection name, put them in your ad names and filter on those in Ads Manager.

Does this work for video ads?

Partly. A video ad is tagged from its thumbnail image, not from its frames, sound or captions, so the tags describe how the video looks at first glance rather than how it plays.

How does pattern mining differ from a creative analytics tool?

A creative analytics tool shows how each ad performed. Bach.ai’s pattern view groups ads by their tag combination and compares each combination with your account’s average ROAS, so it answers which kinds of ad work for you, not just which ad did. Compare the products on their own pages before you choose; the Bach.ai vs Motion comparison walks through one example.

What if my creatives don’t fit the taxonomy?

Most dimensions have an “other” value for creatives that fit none of the listed options, and those ads are still counted. A combination with fewer than 3 ads is not reported as a pattern at all.

Method and sources

“Tagging every asset on consistent dimensions — hook type, format, offer framing, call to action — turns a pile of individual results into a pattern you can query.”

Source: Where this guide describes platform behaviour, it follows Meta’s published advertising and Marketing API documentation, which changes without notice — verify anything load-bearing against the current version before you act on it. Every threshold the guide asks you to supply is first-party, drawn from your own account exports and commerce ledger, because no external benchmark can stand in for your own margin structure.

Sources: Bach.ai product code for creative tagging and pattern analysis (checked 2 Oct 2026).