Cross-platform budget allocation

Why average CPA misleads

Average cost per acquisition is a backward-looking number computed across existing spend. It says nothing about what the next increment of spend would return, which is the actual decision being made.

Two platforms with the same average CPA can behave completely differently at the margin. One may be in its efficient range, where additional spend converts at similar rates. The other may already be saturating, where the next increment converts far worse.

Allocating on averages therefore tends to over-fund the platform that saturated early and under-fund the one with headroom.

The learning-phase floor

Every platform needs a minimum volume of conversions to optimise. Below that floor, campaign performance is dominated by the system’s uncertainty rather than by your targeting or creative.

This is the argument against spreading a small budget across three platforms. Three campaigns each below the learning floor perform worse than one campaign above it, even though the total spend is identical.

The practical rule is to concentrate until each active platform is producing conversions consistently, and only then consider adding another.

How many platforms can you run?

One platform is enough for most small advertisers. Two is the practical maximum for a mid-sized budget. Three or more only makes sense where the budget for each remains above the learning floor, which implies a fairly large total.

The exception is deliberate testing. A small test budget on a new platform is not a permanent allocation decision, and should be sized and time-boxed as an experiment rather than blended into the standing split.

Assigning platforms by intent stage

A useful split is by where the platform sits in the buying journey rather than by which reports the best cost per acquisition.

Search captures demand that already exists. Meta creates demand and works on audiences that have shown interest or resemble those who have. TikTok reaches audiences earlier, where the creative does more of the work of generating interest.

Allocating by intent reduces the tendency to compare platforms on a single metric when they are doing different jobs.

Testing a new platform without wrecking the account

Test with a fixed budget, a fixed duration and a defined success measure decided in advance. The most common failure is a test with no endpoint, which then quietly becomes a permanent underperforming allocation.

Keep the test structurally separate from established campaigns. Mixing a new platform into an existing budget line makes it impossible to tell whether the change helped.

Decide before you start what result would justify scaling and what result would justify stopping. Ambiguity at the start is what produces platforms that run for a year on inertia.

When to cut a platform

Cut when a platform has had a fair test at adequate volume and its marginal return is below the alternative use of that budget.

Distinguish an honest failure from an unfinished test. A platform that never reached the learning floor was not tested; it was underfunded. A platform that reached it and still underperformed is a genuine result.

Cutting cleanly matters too. Reducing to a token budget keeps a campaign alive below its learning floor and produces poor performance indefinitely, which is worse than stopping.

Frequently asked questions