If you’re running a CBO campaign and see:

• one ad set spending everything,
• three ad sets starving,
• no consistent delivery,

…you are experiencing Meta’s predictive spend allocation under the Andromeda system.

This behavior is common in 2025.
It doesn’t mean your campaign is broken.
It means the algorithm has made a decision - and it’s your job to understand why.

This article explains exactly why CBO overspends on one ad set, what it means, and the fixes that actually work.

The Core Truth: CBO Doesn’t Split Your Budget “Fairly” - It Splits It “Predictively”

CBO doesn’t aim for balance.
CBO aims for results.

Under Andromeda, Meta uses early behavioral signals to predict which ad set:

• will get the cheapest conversions,
• has the highest volume potential,
• fits your optimization event best,
• aligns with the current auction environment.

That ad set becomes the “primary route” — and gets most of the budget.

Let’s break down the real causes of uneven spend.

Reason #1 - One Ad Set Has Better Early Signals

This is the single biggest reason.

Meta aggressively weights early performance indicators:

• hook rate
• CTR
• CVR
• time-on-ad interactions
• scroll engagement
• comments, shares, saves
• predicted conversion paths

If one ad set gets even a slight early edge, Meta pours budget into it.

Fix:

• Give each ad set 1–2 new creatives to rebalance signals
• Let the campaign run 48–72 hours before judging
• Avoid “killing” ad sets too early

CBO magnifies small creative differences.

Reason #2 - Audience Overlap Is Causing Internal Competition

If multiple ad sets target overlapping populations, Meta:

• consolidates spend into one
• starves the others
• avoids bidding against itself

This creates the illusion that “CBO chose a winner” when in reality:
your ad sets were competing with each other.

Fix:

• Combine similar interests
• Use broader targeting
• Remove layered exclusions
• Use a single Broad ad set + one more proven interest cluster

The more overlap, the stronger the imbalance.

Reason #3 - Your Non-Spending Ad Sets Have Too Small an Audience

Small audiences:

• exhaust quickly
• have higher CPMs
• are more volatile
• provide fewer signals

Meta deprioritizes them because they aren’t scalable.

Fix:

• Expand audiences
• Combine small interests
• Use 3–5M+ minimum audience size
• Reduce geographic segmentation

Bigger audiences = more spend.

Reason #4 - One Ad Set Entered a High-Quality Pocket

Andromeda is pocket-based:
Meta delivers ads into behavioral clusters of users with shared patterns.

If one ad set finds a “hot pocket,” Meta:

• aggressively exploits it
• deprioritizes all other pockets
• pushes 80–95% of spend to that route

This can happen even if the audiences look identical.

Fix:

• Leave CBO alone when a strong pocket is performing
• Add fresh creatives to non-spending ad sets so they can find new pockets
• Don’t turn off high performers prematurely

Pocket discovery is normal and very powerful.

Reason #5 - Your Ad Sets Have Mixed Intent or Weak Creative Matching

If one ad set has creatives that align better with:

• the targeting
• the demographic
• the platform placement
• the pocket
  —then it becomes Meta’s preferred route.

Weak creative relevance = starvation.

Fix:

• Assign creative clusters to specific ad sets
• Don’t dump unrelated creatives into the same CBO
• Use angle-based testing before entering CBO

Relevant creative → stronger predictive scoring → more spend.

Reason #6 - Your Optimization Event Is Too Expensive

CBO struggles when:

• optimization event CPA is high
• daily budget is low
• conversion density is low

Meta funnels spend into the ad set most likely to gather optimization events quickly.

Fix:

• Temporarily switch to ATC/VC optimisation
• Raise budgets to 3× CPA
• Consolidate campaign structure

CBO prefers predictable conversion data.

Reason #7 - Excessive Editing Reset the Allocation Model

Every CBO change forces recalibration:

• budgets
• audiences
• creatives
• placements
• bid strategy
• attribution settings

Each change resets “confidence scoring.”

Until stability returns, Meta will only aggressively spend in one direction.

Fix:

• Adopt a 72-hour “no edits” rule
• Make batch edits, not reactive ones
• Allow pockets to stabilise

Editing is one of the most common causes of uneven spend.

Reason #8 - Your Weak Ad Sets Never Exited Learning

Learning = low priority.
Optimized = high priority.

If 1 ad set exits learning, and 3 others are stuck in learning, CBO treats the optimized ad set as the only “safe” route.

Fix:

• Increase overall budget
• Fix pixel or funnel issues
• Add more stable creatives
• Remove overly-narrow audiences

Learning completion drives spend allocation.

How to Fix Uneven Spend (The 2025 Protocol)

This is your step-by-step method to rebalance delivery:

Step 1 - Add 2–4 Fresh Creatives to Starved Ad Sets

This is the fastest way to rebalance signals.

Step 2 - Consolidate Overlapping Audiences

Fewer, larger surfaces → more stable allocation.

Step 3 - Leave CBO Alone for 48–72 Hours

Do not touch it.
Do not throttle it.
Give it time.

Step 4 - Remove Bad Ad Sets (Not Starved Ones)

Only kill an ad set if:

• CTR is terrible
• hook rate is poor
• CPM is significantly higher
• CPA is wildly uncompetitive

“Not spending” ≠ “bad.”

Step 5 - Promote Tight Creative Clusters

Ad sets perform better when creatives share:

• angle
• messaging
• visual style
• hook theme

Clustered signals improve allocation.

Step 6 - Use ABO for Audience or Creative Testing

Then promote winners into CBO.

When You Should Not Fix It

If the overspending ad set:

• has the best CPA
• has stable ROAS
• has good CTR
• is scaling predictably

…then do nothing.

CBO is doing its job.

Final Thoughts

Uneven spend isn’t a glitch - it’s Meta’s prediction engine doing the heavy lifting.
Your role is to create an environment where multiple ad sets can perform well enough to earn spend.

Do that, and CBO becomes one of the most powerful scaling tools in 2025.