Campaign architecture
Account structure, budget allocation, bid strategy and the learning phase.
Campaign structure is not an organizational preference. It decides how concentrated the signal you hand the algorithm is. Splitting the same budget across many small ad sets dilutes that signal, and dilution is a direct performance loss.
The consolidation principle
Optimization models need statistically meaningful volume. When you split the budget, each fragment tries to learn on its own and none of them gets enough data. The result is slow learning everywhere, volatile results, and higher cost.
A reasonable skeleton for most accounts:
| Campaign | Purpose | Budget share |
|---|---|---|
| Broad acquisition | New customers, automated targeting | 60–75% |
| Retargeting | Cart and checkout abandoners | 10–20% |
| Testing | New creative concepts | 10–20% |
Keep the number of active ad sets as low as you can. The only legitimate reason to open a separate ad set is that you must control that segment's budget separately — "a different audience" is not on its own a reason.
The learning phase
Every new or significantly edited ad set enters a period where the model tries to separate patterns from noise. During it, costs are volatile and usually above target.
The practical threshold: roughly 50 optimization events per ad set per week. Below that, ad sets get stuck in a "learning limited" state and never produce consistent results.
The minimum weekly budget for an ad set to produce 50 conversions:
minimum weekly budget ≈ target CPA × 50
If your CPA is $20, an ad set needs at least $1,000 a week. If your budget cannot support splitting that across three ad sets, do not open three ad sets.
If you cannot reach 50 events, you have two options: consolidate budget into one ad set, or move to a shallower optimization event (initiate checkout, add to cart). The second speeds up learning but lowers signal quality — treat it as a temporary fix.
Changes that reset learning
| Change | Resets learning? |
|---|---|
| Budget change over 20% | Yes |
| Bid strategy or target CPA change | Yes |
| Optimization event change | Yes |
| Targeting or placement change | Yes |
| Adding new creative to the set | Usually no |
| Turning off an existing creative | Usually no |
The working discipline that follows: batch your changes, apply them in one session, then leave it alone. An account that gets small daily tweaks lives in a permanent learning phase.
Choosing a bid strategy
| Strategy | What it does | When to use it |
|---|---|---|
| Lowest cost | Spends the budget, tries to minimize cost | Volume first, margin comfortable |
| Cost cap | Keeps average CPA under a set ceiling | Tight margin, CPA discipline required |
| Bid cap | Hard ceiling on auction bids | Experienced operator, strict target |
| Target return (ROAS) | Optimizes on value | Highly variable basket size |
Capped strategies buy predictability but can choke delivery. Set the cap below market price and the campaign simply will not spend — the result is not "cheap", it is "nothing". Start caps 15–25% above your actual current CPA.
Budget increase cadence
Doubling a budget overnight resets learning and usually moves cost up permanently. Stepping up preserves the signal and leaves you a way back.
| Step | Practice |
|---|---|
| Increment | 20–30% at a time |
| Wait | 2–3 days before the next increase |
| Stop condition | CPA runs 20% above target — stop increasing |
| Rollback | Return to the previous budget level, wait out re-learning |
If you need to grow faster, the path is more creative volume, not a harder push on budget. What caps your budget is usually not money — it is the number of winning creatives you have.