CBO hands budget allocation to the algorithm. Whether that is smart depends entirely on the quality of the data it is allocating against.
What CBO actually does
Instead of setting budget per ad set, you set it at campaign level and Meta distributes in real time toward whatever is performing.
On one donation campaign, a single ad absorbed roughly 90% of budget — ₹23,611.74 ≈ $243.84— while the other three received ₹1,459.30≈ $15.07, ₹639.17 ≈ $6.6and ₹541.87≈ $5.6. That was correct: the winner outperformed on every metric, and the algorithm found it faster than a human reviewing dashboards weekly would have.
Use CBO when
Your conversion tracking is complete. This is the precondition, not a nice-to-have. CBO allocates toward reported conversions — if reporting is broken, it allocates toward the wrong thing efficiently.
You have enough conversion volume. Roughly 30–50 events weekly per campaign. Below that, the algorithm is making decisions on noise.
Your ad sets are genuinely comparable. Similar audience sizes, similar offers. Mixing a broad audience with a tiny lookalike means the broad one wins on volume regardless of quality.
You are scaling something proven. CBO is a scaling tool, not a discovery tool.
Intent clusters in time windows. Donation and event-driven campaigns often see spikes; CBO captures them where rigid ad set budgets miss them.
Do not use CBO when
You are testing. Testing requires equal spend across variants to produce a fair comparison. CBO will starve a variant before it has enough data to judge, and you will conclude something false about it.
Tracking is incomplete. Fix that first. Everything else is downstream.
Ad sets have very different audience sizes. The algorithm will favour reach over quality.
You need guaranteed spend on a specific segment. If a particular audience must receive budget for strategic reasons, control it at ad set level.
Advantage+ specifically
Advantage+ campaigns automate targeting, placements, and often creative combinations. They perform well with clean conversion data and a decent creative pool, and they are opaque — you see less about why something worked.
The practical position: excellent for scaling proven offers with good tracking, poor as a substitute for understanding your audience.
Be wary of any agency running everything through Advantage+ and reporting only the headline number. It is convenient for them and uninformative for you.
The sequence that works
Phase 1 — Test with ad set budgets. Equal spend, deliberate variants, one hypothesis at a time. Accept that some tests will cost more per result; you are buying information. On the campaign referenced above, tests ran at ₹193.07 ≈ $1.99and ₹242.34 ≈ $2.5before the winner scaled at ₹85.48≈ $0.88.
Phase 2 — Identify the winner on cost per result and on lead quality, not on volume alone.
Phase 3 — Move to CBO with the winning structure and let allocation automate.
Phase 4 — Scale in steps. Sharp budget increases reset learning. On that campaign, daily budget ran at ₹15,000 ≈ $154.91with maximum daily spend around ₹26,250≈ $271.08.
The mistake this prevents
Manually shifting budget to yesterday's winner is slower than the auction changes. By the time you have reviewed the dashboard and adjusted, conditions have moved.
The point of CBO is not laziness. It is that structural decisions belong to you and allocation decisions belong to the machine — provided you have given the machine accurate data to decide with.