Learning Phase Limited for 14+ days. ROAS dropping. Every edit resets the clock. If this sounds familiar, you are running into one of the most common - and most misdiagnosed - Meta Ads issues in 2026.
This guide covers the 8 causes of stuck Learning Phase, how to diagnose which one you are hitting, and the fix order. Updated for the April 2026 Andromeda rollout that has tightened Learning reset thresholds across most accounts.
What Learning Phase actually means
Note
Learning Phase definition. Meta's Learning Phase is the exploration period when a new or recently-edited ad set is gathering data to optimize delivery. An ad set exits Learning after 50 conversions in 7 days for the optimization event. If it cannot reach 50 conversions in 7 days, it enters 'Learning Limited' - a degraded performance state where Meta's algorithm cannot fully optimize.
The 50-in-7-days threshold has not changed since 2018. What HAS changed in 2026: which edits trigger a Learning reset. Andromeda made this stricter.
8 causes of stuck Learning Phase
Cause 1: Not enough conversion volume
The root cause for most accounts. If your ad set is generating fewer than 50 conversions per week, it cannot exit Learning - that is the floor, not a bug.
Diagnose. Daily budget × 7 ÷ CPA = expected weekly conversions. Below 50, you are structurally Limited.
Cause 2: Optimization event too rare
You are optimizing for Purchase but only get 12 purchases per week. The volume threshold is for the optimization event specifically, not all conversions combined.
Diagnose. Open the ad set. Note the optimization event. Pull the last 14 days of that event count from Events Manager.
Cause 3: Edits triggering Learning reset
Common in 2026. Even small edits (budget +/- 20%, audience tweaks, creative swaps) can reset Learning. Andromeda tightened this threshold around February-March 2026.
Diagnose. Open the ad set's history. If you see Learning reset 2+ times in the last 30 days, you are editing too often.
Cause 4: Audience too narrow
Audience under 1M with low conversion rate cannot generate the volume needed.
Diagnose. Open ad set > audience size estimate. Below 500K, Learning will struggle. Above 5M, audience is fine; problem is elsewhere.
Cause 5: Daily budget too low
Meta's algorithm allocates exploration budget proportional to daily spend. Below 50 USD daily, exploration is constrained.
Diagnose. If daily budget is below 50 USD on a CBO campaign, Learning may never finish. Bump or consolidate ad sets.
Cause 6: Andromeda recent regression (Feb-May 2026)
Specific to right now. Andromeda's personalization model is re-evaluating audience signals. Some accounts see Learning take 2-3x longer to complete than pre-Andromeda.
Diagnose. Started in Feb-March; gets worse around audience changes. Check whether ad sets that previously exited Learning in 5-7 days now take 14+.
Cause 7: Wrong attribution window
If your attribution window is set to 1-day click and your customer journey is 5-7 days, conversions are undercounted toward the 50-conversion threshold.
Diagnose. Open campaign > Attribution Setting. If 1-day click and your CRM shows lag time >1 day, switch to 7-day click + 1-day view.
Cause 8: Frequency capped causing under-delivery
Less common but real. Frequency caps below 1.5 over 7 days suppress delivery to under-50-conversions territory.
Diagnose. Open ad set > Frequency Cap. If set, remove temporarily and observe.
How to fix it: diagnostic order
The Learning Phase fix order
Any account with Learning Limited ad sets
- Best for
- Any account with Learning Limited ad sets
- Pricing
- Free
Pros
- Step 1: Pull weekly conversion volume on optimization event. If below 50, you have a volume problem; consolidate ad sets.
- Step 2: Check edits in last 14 days. If 2+ edits, freeze edits for 7 days and observe.
- Step 3: Check audience size; below 500K, broaden to 1-3M.
- Step 4: Check daily budget; below 50 USD, bump or consolidate.
- Step 5: Verify attribution window matches your real conversion lag.
- Step 6: If Andromeda timing is the cause, give it 14-21 days before further intervention.
- Step 7: Consider switching optimization event to a higher-volume upstream event (Initiate Checkout vs Purchase) for high-funnel ad sets.
Cons
- Patience required - resisting the urge to edit while observing is the hardest part
What to do if you cannot escape Learning Limited
Sometimes the math does not work and Learning Limited is not a bug, it is a structural reality. Three options:
Option 1: Consolidate
Three ad sets each getting 20 conversions/week become one ad set getting 60. Meta's Advantage+ Audience effectively does this automatically.
Option 2: Optimize for an upstream event
Switch from Purchase to Initiate Checkout, or from Lead to Add to Cart. The upstream event has higher volume; Learning exits; Meta still drives toward purchase as a downstream optimization signal.
Option 3: Accept Learning Limited
If consolidation breaks targeting and upstream optimization breaks lead quality, sometimes Limited delivery is the right tradeoff. Performance still happens; it is just less optimized than full Learning Out would be.
When to consolidate vs accept Learning Limited
When this fits
Recommended: Consolidate if: ad sets target overlapping audiences. Multiple ad sets with same creative. Brand-vs-non-brand split that Andromeda can handle automatically. Optimize for upstream event if: optimization event volume is structural floor. Brand has long consideration cycle. Lead-gen with form-fills as Mid-funnel signal.
When to skip
Recommended: Adding more ad sets when Learning is already Limited (split a small pie smaller). Editing daily looking for the bug (the bug is volume, not creative). Switching attribution windows weekly (each switch resets Learning). Cutting budgets on Learning Limited ad sets - usually makes Learning worse, not better.
Autonomous marketing
Grow your business faster with AI agents
- Automates Google, Meta + 5 more platforms
- Handles your SEO end to end
- Improves website conversions
- Runs social media for you
How Hyper helps
Hyper agents monitor Learning Phase status across every ad set continuously. When Learning resets unnecessarily, when conversion volume falls below the 50/week threshold, when consolidation would help - the agent flags it and recommends the action. Many of our customers have not had a Learning Limited ad set for months because the agent caught the issue and fixed it before it became a Learning regression. Case study: /blog/ai-marketing-case-study.
Autonomous marketing
Grow your business faster with AI agents
- Automates Google, Meta + 5 more platforms
- Handles your SEO end to end
- Improves website conversions
- Runs social media for you
Frequently asked questions
Q: How long does Meta Ads Learning Phase last?
Up to 7 days from creation or significant edit, requiring 50 conversions on the optimization event. Below that volume the ad set enters Learning Limited and never fully exits. Andromeda has stretched typical exit time from 4-7 days to 7-14 days for many accounts in 2026.
Q: Will my ad set exit Learning if I leave it alone?
Only if it generates 50 conversions on the optimization event in any 7-day rolling window. Without that volume, no amount of patience exits Learning. The fix is volume, not waiting.
Q: Does Learning Phase reset count as performance loss?
Yes. Each reset costs roughly 5-15% of the next-week ROAS as the algorithm re-explores. 3 resets in a month can drag account-wide ROAS down 10-20%. Bundle edits to minimize reset count.
Q: Should I optimize for Purchase or Initiate Checkout?
Optimize for Purchase if you generate 50+ purchases per ad set per week. If under that, optimize for Initiate Checkout (higher volume, exits Learning) and let Meta's algorithm drive downstream to purchase. Most accounts with under 50 purchases/week get better blended ROAS optimizing upstream.