Meta Ads: when campaign budget optimisation beats setting budgets yourself

Campaign budget optimisation lets Meta move money between ad sets. Ad set budget optimisation keeps you in control. Teams argue about which is superior; the honest answer is that the right one depends on whether you are trying to learn something or trying to scale something you already learned.

Use ABO when you are testing

If you want a clean read on which audience or creative works, you need each ad set to get comparable spend. CBO will not do that — it will push budget to the early leader, which is usually the ad set that got a cheap conversion in the first hour, not the one that will perform over a month. For tests, fix the budgets and let each cell get enough impressions to be believable.

Use CBO when you are scaling

Once you know which creative and which audience work, CBO does exactly what you want: it finds the cheapest conversions across the ad sets you have already validated. This is where hand-managed budgets lose, because no human reallocates budget hourly.

The structure that works for most accounts

• One testing campaign on ABO with a fixed, modest budget. New creatives enter here.
• One scaling campaign on CBO holding only proven creative and audiences.
• Graduate a creative from testing to scaling on a threshold you set in advance — a cost per result and a minimum number of results — not on a hunch.
• Retire creatives on frequency and on declining click-through, not on how bored the team is of them.

Things that quietly break both

• Too many ad sets. Every split fragments the conversion data the algorithm needs. Broader audiences with strong creative usually beat narrow slicing.
• Editing live ad sets. Significant edits restart learning. Duplicate instead.
• Optimising for the wrong event. If purchase volume is too low to learn from, optimise for the highest-value event you can generate enough of, then move up.
• Ignoring creative fatigue in a CBO campaign, where budget concentration accelerates burn.

Structure the account so testing and scaling never share a campaign, and the CBO-versus-ABO argument stops mattering — you are simply using each for the job it is good at.