Every paid campaign divides money across three jobs:
Prospecting — reaching people who don't know you. Retargeting — reaching people who already engaged. Testing — finding out what works next.
The split between them matters more than most of the settings people spend their time on, and the conventional wisdom about it is out of date.
Why the old advice is stale
For years the standard was a fairly even funnel split — something close to equal thirds across top, middle and bottom, matching the TOF/MOF/BOF diagram everyone had on a slide.
Two things broke it.
Automated broad targeting got better. Platform delivery systems now find responsive people inside a large audience more effectively than manual interest stacking does. Feeding a broad prospecting audience a large budget produces better results than it used to, because the system has more room to work.
Signal loss hurt retargeting. Tracking restrictions shrank retargeting audiences substantially. The pool is smaller and less complete than it was, so the same percentage of budget chases fewer people — and over-funding a small audience just means showing the same ads repeatedly to people already tired of them.
The net effect is that budget has shifted towards broad prospecting and away from heavy retargeting.
A modern default
A reasonable starting point in the current era:
| Job | Share | Why |
|---|---|---|
| Prospecting | ~75% | Broad, algorithm-led. Where growth comes from and where the creative is really tested. |
| Retargeting | ~15% | Smaller audience, higher intent. Enough to cover it without saturating. |
| Testing | ~10% | Ring-fenced, deliberately spent on learning rather than returns. |
Treat this as a default, not a law. It's the split Acumin's Ad Lab starts from, and it reflects the consolidated, broad-prospecting-heavy structure that current platform delivery rewards.
When to move off the default
More prospecting when: you're new or entering a new market, your retargeting pool is small, you're growing, or your creative is working and the constraint is reach.
More retargeting when: you have meaningful traffic, a considered purchase with a long decision cycle, or a genuinely different bottom-of-funnel offer worth showing.
More testing when: you don't know what works, you're entering a new category, or your current creative is fatiguing. A third of budget on testing is defensible when you're genuinely lost — as long as you call it that.
Less testing when: budgets are so small that a 10% slice can't produce a readable result. Below a certain scale it's better to run sequential tests with the whole budget than a parallel test nobody can read.
The floor that constrains everything
Before any split, check that each ad set can accumulate enough data to mean something.
Platform delivery needs a volume of optimisation events before it stabilises — the working figure on Meta is around 50 events within 7 days per ad set, documented in Meta's own Ads Help Center, which is where you should check it since the specifics change.
The arithmetic:
daily budget × 7 ÷ estimated cost per event = events per weekIf your split produces ad sets that can't clear that, the split is wrong regardless of how sensible the percentages look. Fewer, better-fed ad sets beat a structurally elegant plan that starves all of them.
That's the most useful thing in this article for small advertisers: below a certain spend, the correct number of ad sets is one.
Judge each job on its own metric
The other common failure is applying one success measure across all three.
Prospecting should be judged on reach into the right audience and cost per new engaged person. Judged on immediate conversions it will always look like the worst-performing part of your account — and it's what makes everything else possible.
Retargeting looks efficient by design, because it's converting demand that already existed. Its apparent return is partly borrowed from prospecting's work. Scaling it because the numbers look good is the most common misallocation in paid social.
Testing should be judged on what you learned, not on returns. A test that conclusively kills an expensive idea was a good use of money.
Applying one metric across all three reliably leads to defunding prospecting and over-funding retargeting, which works for a quarter and then stops working.
Changing the split
Move it slowly. Large reallocations disrupt delivery and reset learning. Move 10–15% at a time and let it stabilise.
Change one thing. If you change the split, the creative and the targeting in the same week, you've learned nothing.
Give it long enough. At minimum, past the learning phase. Two weeks is a reasonable floor for most budgets.
Reallocate on observed efficiency, not on a forecast. Compare campaigns that have actually recorded results, and require a meaningful gap before you move money — a leader that beats a laggard by a few percent is inside the noise.
Where Acumin fits
Ad Lab starts from a 75/15/10 prospecting-heavy split, allocates the daily budget across the stages, and the budget planner runs the learning-phase arithmetic above from your daily spend and estimated cost per event — telling you whether the structure can plausibly clear it before you launch. Those are bands computed from your inputs, never a promised result.
Budget comparison across campaigns is observed only: it compares campaigns that have recorded results, ranks them by real cost per result, and suggests reallocation only on a meaningful gap — the leader has to beat the laggard by at least 15% before it says anything. No projection, no causal claim, no invented budget.
The results themselves come from Meta's own reporting today, entered against the campaign. Automatic read-back is pending Meta's app review, and it's labelled that way in the product.
How to use this today
Take your current daily budget and estimated cost per result, and run the division for each ad set you're running.
Count how many clear 50 events a week. If the answer is none, consolidate — that one change will do more than any percentage adjustment.
Related: Structuring a Meta campaign from an organic winner is the build this split sits inside. Reading ad results without fooling yourself is how to judge whether the split worked.