Budgeting: Campaign Budget Optimization vs. Ad Set budgets

Part 4 flagged that budget can live at either the campaign level (Campaign Budget Optimization) or the ad set level. This part covers what CBO actually does with a shared budget, and the real cases where ad-set-level budgets still make more sense.

How CBO actually works

Campaign: Sales — Coffee Subscriptions (CBO: On, $30/day total)
  Ad Set: Coffee Enthusiasts        → received $19/day today
  Ad Set: Gift Shoppers              → received $7/day today
  Ad Set: Retargeting — Cart Abandoners → received $4/day today

With CBO on, one shared budget is set at the campaign level, and Meta's delivery system continuously redistributes it across every ad set beneath it, in real time, based on which one it currently predicts will produce the best results for that specific dollar. This allocation genuinely shifts day to day — a strong-performing ad set can receive a meaningfully larger share on one day and less the next, as real delivery data updates the prediction.

When CBO is the right default

For most accounts, most of the time, CBO produces better overall results than manually guessing budget splits — the delivery system has direct, continuously updated visibility into real performance differences between ad sets that a human reviewing daily or weekly reports simply can't react to as quickly or precisely.

When ad-set-level budgets still make sense

Ad Set: Coffee Enthusiasts        → manual budget: $20/day (guaranteed minimum)
Ad Set: Small Test — New Interest  → manual budget: $5/day (deliberately capped)

CBO can starve a genuinely important but currently lower-performing ad set of budget entirely, in favor of whatever's predicting best results right now — which is a real problem for a small-budget test ad set (part 5's "detailed targeting expansion" testing, for instance) that needs a guaranteed minimum spend to gather enough data to actually evaluate, regardless of its early performance relative to an already-proven ad set. Manual ad-set budgets guarantee each one gets its allocated share, useful specifically when deliberate, controlled testing matters more than pure short-term efficiency.

The learning phase: a real, documented delivery state

"Learning" status: shown directly on an ad set in Ads Manager

A new ad set (or one with a significant recent edit) enters a "learning phase" — Meta's system actively exploring delivery options with less certainty than it'll have later, generally until it accumulates around 50 optimization events (the specific conversion the ad set is optimized for). Performance during this phase is genuinely less stable and less representative of the ad set's eventual real performance, which is exactly why part 2's guidance to avoid frequent early edits matters — every significant edit can reset the learning phase, delaying the point at which delivery actually stabilizes.

Why frequent budget or targeting edits hurt more than they seem to

Editing an ad set's budget by more than ~20% in a single change,
or editing targeting, creative, or bid strategy meaningfully,
can restart the learning phase

This is a genuinely counterintuitive real cost: reacting too quickly to a rough first few days by making a significant change can reset the exact learning process needed to reach stable, efficient delivery — compounding the instability rather than fixing it. A smaller, more gradual budget change generally avoids retriggering a full learning phase reset; a dramatic one usually does.

A practical budget-change discipline

Increase budget by 20-30% at a time, not more, spaced at least a few days apart

This specific, real guideline — a moderate, gradual increase rather than a dramatic jump — is what lets a campaign scale spend without repeatedly forcing itself back into an unstable learning phase, a direct practical application of the mechanism just described.

Next: bidding strategies — Lowest Cost, Cost Cap, and Bid Cap — and how they interact with the learning phase and CBO budget decisions just covered.