How the Meta Learning Phase Works (And How to Stop Breaking It) (2026)

TL;DR

Meta's learning phase requires approximately [50 conversion events per week](https://www.facebook.com/business/help/112167992830700) per campaign to complete. During this period, delivery is unstable and CPA runs 20-50% higher than post-learning performance
The most common way brands break the learning phase is making changes too frequently. Every significant edit to budget, creative, audience, or optimization resets the counter, and a campaign stuck in perpetual learning never optimizes
Budget changes under 20% and adding creative to a healthy campaign with 8+ active ads generally don't reset learning. Large budget swings, audience changes, and optimization event changes almost always do
If your campaign shows "Learning Limited," you don't have enough conversion volume at your current budget. The fix is consolidation (fewer campaigns, broader targeting), not more spend into an underperforming structure

What is the Meta learning phase?

When you launch a new campaign or make a significant change to an existing one, Meta’s algorithm enters the learning phase. During this period, the delivery system is actively exploring which users to show your ads to and how to bid for them. It needs data to build a reliable model, and that data comes from conversions.

Meta requires approximately 50 conversion events per week per campaign to complete the learning phase. At a $30 CPA, that’s $1,500/week in spend. At a $50 CPA, that’s $2,500/week. The conversion event must match your optimization objective, so if you’re optimizing for purchases, Meta counts purchases, not add-to-carts or page views. For the full strategic framework, see our Meta Ads for eCommerce: The Complete Guide.

During learning, expect higher CPAs and less stable delivery. The algorithm is testing different audience segments, placements, and bid levels. Some impressions will go to users who don’t convert because the algorithm is still mapping the conversion landscape. This is normal. It’s not a sign that your ads aren’t working. It’s the algorithm doing its job.

The learning phase typically takes 7-14 days for campaigns with sufficient budget. You’ll see “Learning” in the Delivery column of Ads Manager. Once the campaign exits learning, it shows “Active” and delivery stabilizes.

Our finding: Across our managed accounts, the brands that see the best post-learning performance are the ones that resist the urge to make changes during the first 7-10 days. CPA during learning is almost always higher than where the campaign settles, but the degree of improvement varies widely by account and category. The consistent pattern: every account we’ve onboarded that was “stuck” with high CPAs had the same root cause — constant changes that kept campaigns in perpetual learning.

What resets the learning phase?

Not every change resets learning. Meta distinguishes between significant edits (which reset the counter) and minor adjustments (which don’t). Understanding the difference prevents accidental resets.

Changes that reset the learning phase:

Budget changes above 20%. Increasing daily budget from $100 to $130 is usually fine. Jumping from $100 to $200 resets learning because the algorithm needs to find an entirely new set of users to fill the doubled delivery volume
Adding or removing ads in small campaigns. In a campaign with 3-4 active ads, adding or removing one changes the creative pool significantly. The algorithm needs to relearn delivery distribution. In larger campaigns (8+ ads), adding a single ad is less disruptive
Changing the optimization event. Switching from “Purchase” to “Add to Cart” or vice versa completely resets the model because the algorithm is learning a different conversion behavior
Changing targeting. Modifying audience definitions (adding or removing interests, changing lookalike percentages, adjusting age/gender) changes who the algorithm delivers to. It needs to rebuild its user model from scratch
Pausing and restarting. Pausing a campaign for more than a few hours effectively resets learning. The algorithm’s model degrades during inactivity, and restarting triggers a new exploration phase
Bid strategy changes. Switching from lowest cost to cost cap or target ROAS changes how the algorithm bids. Each strategy has a different optimization model

Changes that generally don’t reset learning:

Budget increases under 20%
Updating ad copy or headlines (without changing the visual creative)
Adjusting the existing customer budget cap in ASC by small increments
Changing the campaign name
Adjusting the ad schedule within the same timezone

The gray zone:

Adding 1-2 new ads to a campaign with 8+ existing ads usually doesn’t fully reset learning, but it does cause a brief period of redistribution while the algorithm evaluates the new creative against the existing pool. Expect 24-48 hours of slightly elevated CPA. This is why graduating winners from your testing campaign into a healthy ASC with 8+ active ads is generally safe. See our creative testing system guide for the graduation process.

What does “Learning Limited” mean?

“Learning Limited” is different from “Learning.” It means the campaign can’t get enough conversions to complete the learning phase at the current budget and structure.

The cause is almost always insufficient conversion volume. Your campaign isn’t generating 50 conversions per week because:

Budget is too low for your CPA. At a $40 CPA, you need $2,000/week ($285/day) per campaign to hit 50 conversions. If your daily budget is $100, the math doesn’t work
Too many campaigns splitting your budget. Five campaigns at $50/day each is worse than two campaigns at $125/day. The algorithm performs better with concentrated data than fragmented data
Targeting is too narrow. Small audiences limit the number of people the algorithm can show your ads to, which limits conversions. Broader targeting gives the algorithm more room to find buyers
Optimization event is too rare. If you’re optimizing for purchases but only getting 5-10 per week, consider optimizing for a higher-volume event temporarily (like add-to-cart) to give the algorithm more learning signal. Switch back to purchases once you have the volume to support it

The fix is consolidation, not more budget.

Before increasing spend on a Learning Limited campaign, restructure:

Reduce campaign count. Merge similar campaigns. One ASC campaign with broad targeting performs better than three manual campaigns with segmented audiences at low budgets
Broaden targeting. Remove interest restrictions. Enable Advantage+ Audience on manual campaigns. Let the algorithm find buyers instead of constraining it. See our Advantage+ Audience guide for when to enable this
Consolidate optimization events. If you’re running some campaigns on purchases and others on add-to-cart, pick one event and consolidate
Only then increase budget. Once you’ve consolidated, increase budget using the 20% rule to avoid resetting learning on the restructured campaign

For budget minimums by growth stage, see our budget guide.

Our finding: “Learning Limited” is the most misunderstood status in Ads Manager. Most brands see it and assume they need to spend more. But in 60-70% of cases we audit, the fix isn’t more budget. It’s fewer campaigns. A brand running four campaigns at $75/day each is spending $300/day total, enough for 50 conversions per week at a $42 CPA. But each individual campaign is only getting 12-13 conversions per week. Consolidating into two campaigns at $150/day each solves the Learning Limited status without spending an additional dollar.

How do you protect the learning phase?

The best learning phase strategy is prevention. Structure your account so that learning resets happen as rarely as possible.

The one-change-at-a-time rule. Never stack multiple changes in the same week. If you’re adding new creative, don’t also change the budget. If you’re increasing budget, don’t also adjust targeting. Each change creates a variable the algorithm needs to learn. Multiple simultaneous changes create compounding uncertainty that extends the learning period significantly.

The 20% budget rule. When scaling, increase budget by no more than 20% every 3-4 days. This gradual approach lets the algorithm expand delivery without resetting. A $500/day campaign reaches $1,000/day in about 2 weeks. It feels slow, but it preserves the optimization data the algorithm has already built. For the full scaling framework, see our scaling guide.

Separate testing from scaling. New creative should never enter your ASC scaling campaign directly. It enters your testing campaign first, gets evaluated using the 5x CPA rule, and only graduates to ASC if it proves itself. This protects your scaling campaign’s learning stability. See our creative testing system guide for the full methodology.

Separate seasonal from evergreen. Run Black Friday, holiday, and product launch campaigns in their own manual ABO campaigns with separate budgets. When the event ends, pause the seasonal campaign. Your evergreen ASC continues undisturbed. See our Advantage+ vs. manual campaigns guide for when to use each campaign type.

Don’t panic during learning. The first 3-5 days of a new campaign will show elevated CPA. This is expected. The worst response is cutting budget or pausing ads during this period, because that either resets learning or kills the campaign before the algorithm has a chance to optimize. If you’ve set up the campaign correctly (right budget, right optimization event, broad targeting, strong creative), give it 7-10 days before evaluating.

Our finding: The single most damaging pattern we see in eCommerce Meta accounts is what we call “the optimization spiral.” A brand launches a campaign. CPA is high during learning. They reduce budget on day 3. Learning resets. CPA stays high. They change the targeting on day 5. Learning resets again. By day 10, the campaign has been in perpetual learning and never had a chance to optimize. They conclude “Meta doesn’t work for our brand” and pause everything. When we rebuild these accounts with proper structure and the patience to let learning complete, performance improves within 2-3 weeks without changing the creative or the product.

How does the learning phase differ across campaign types?

Different campaign types handle learning differently. Understanding these differences prevents mistakes when managing a multi-campaign architecture.

ASC (Advantage+ Shopping Campaigns): ASC runs a single campaign with all creative in one pool. Learning applies to the campaign as a whole, not individual ads. Adding a new ad to a healthy ASC with 8+ active ads causes minimal disruption. The algorithm briefly redistributes delivery, then settles. Removing a top-spending ad has more impact because the algorithm needs to reallocate that spend. See our ASC playbook for managing creative inside ASC.

Manual CBO campaigns: Learning applies at the campaign level, but budget distribution happens at the ad set level. Each ad set effectively has its own micro-learning. Campaigns with multiple small ad sets fragment learning across all of them. Consolidating into fewer, larger ad sets accelerates learning.

Manual ABO campaigns: Learning applies at the ad set level because each ad set has its own fixed budget. This means each ad set needs to independently generate enough conversions to exit learning. ABO is appropriate for seasonal campaigns where you want precise budget control, but it’s less efficient for evergreen campaigns because of the fragmented learning.

Retargeting campaigns: Retargeting to small, warm audiences (site visitors, cart abandoners) may not hit 50 conversions per week. This is acceptable. Retargeting audiences are already high-intent, so the algorithm needs less exploration data to deliver efficiently. A retargeting campaign in “Learning Limited” can still perform well because the audience quality compensates for the limited data volume. See our retargeting guide for more detail.

Frequently Asked Questions

How long does the learning phase last?

Typically 7-14 days for campaigns with sufficient budget. The timeline depends on conversion volume, not calendar days. A campaign generating 10 conversions per day exits learning in about 5 days. A campaign generating 2 per day takes 3-4 weeks. Budget that generates 50 conversions per week at your CPA is the threshold for completing learning in a reasonable timeframe.

Should I pause a campaign that’s in the learning phase?

No. Pausing resets learning and wastes the data the algorithm has already collected. If CPA is high during learning, that’s expected. The only reason to pause during learning is if something is fundamentally wrong (broken tracking, wrong optimization event, wrong landing page). For diagnosing actual performance issues, see our ROAS diagnosis guide.

Does adding one new ad to ASC reset the learning phase?

Generally no, if the campaign has 8+ active ads. The algorithm briefly redistributes delivery (24-48 hours of mild CPA fluctuation) but doesn’t fully reset. Adding multiple ads simultaneously or adding to a small campaign (3-4 ads) has more impact. This is why we graduate proven winners one at a time from testing to ASC.

What conversion event should I optimize for during learning?

Optimize for purchases whenever possible. Higher-funnel events (add-to-cart, initiate checkout) produce more conversion volume, which exits learning faster, but they also train the algorithm to find users who add to cart rather than users who buy. If you genuinely can’t generate 50 purchase conversions per week, use add-to-cart as a temporary optimization event, then switch to purchases once volume supports it.

Is the 50-conversion threshold per campaign or per ad set?

Per campaign for CBO and ASC campaigns (since budget is managed at the campaign level). Per ad set for ABO campaigns (since each ad set has its own budget). This is one reason CBO and ASC are preferred for evergreen campaigns: they consolidate conversion data at the campaign level, making it easier to hit the 50-conversion threshold.