Metrics and analysis
How to split budget across Meta, Google and TikTok
Splitting by intuition is expensive, and splitting by each platform CPA is too. The criterion that works, and why the cheapest channel does not always deserve more.
Also available in: Português · Español
Three channels, one budget, and the question comes back every planning meeting: how much in each?
The two most common answers are wrong. "Split it evenly" ignores that the channels do different jobs. "Put it all in the lowest CPA" ignores that the lowest CPA usually has the lowest ceiling.
Why the lowest CPA misleads
The channels are not competing for the same thing.
Search harvests demand that already exists. Someone searching "insurance quote price" has already decided they want it. The ad only needs to be there. Low CPA, low ceiling: the number of people searching per month does not change because you raised the budget.
Social creates demand. Nobody opens Instagram wanting to buy anything. The ad has to generate the interest before capturing it. Higher CPA, far higher ceiling.
Putting all the budget into search because its CPA is lower works until the month search saturates — and then its CPA rises, because you are buying increasingly irrelevant terms, and you do not have the demand-creation channel running to feed next month's search.
The two feed each other. A good share of the people searching your brand on Google saw an Instagram ad first. If you cut social, brand search drops a month later — and the report will record it as if search got worse on its own.
The common basis: the CRM
Comparing channels in each platform's dashboard is comparing different rulers. Each platform uses its own window, its own model, and claims what it can.
The only common basis is the sale recorded in the CRM with the right source:
| channel | spend | sales (CRM) | revenue | cost/sale | return |
|---|---|---|---|---|---|
| $12,000 | 24 | $168,000 | $500 | 14.0x | |
| Meta | $28,000 | 41 | $287,000 | $683 | 10.3x |
| TikTok | $6,000 | 5 | $35,000 | $1,200 | 5.8x |
That table exists in no ads manager. It requires the source reaching the CRM, the work described in connecting Meta Ads to your CRM.
And it has to come with coverage beside it: if 30% of sales have no source, the budget split is being decided on 70% of reality.
The decision criterion
First: is each channel above the minimum?
A channel with too little budget does not produce enough data to be judged. $6,000 in a channel whose cost per sale is $1,200 yields five sales a month — pure noise.
Either give the channel enough budget to be assessable, or switch it off. Keeping a channel on a homeopathic dose for a year is paying not to learn.
Second: where is the ceiling?
A channel returning 14x and saturated will not take more budget — raising it there only worsens its CPA. A channel returning 10x with room will.
The saturation signal in search is impression share lost to budget approaching zero: you are already showing up for everything worth showing up for. In social, it is frequency rising without reach rising.
Third: what is each one's job?
Split the budget by function, not only by performance:
| function | how much | what happens without it |
|---|---|---|
| Harvest existing demand | whatever the channel absorbs without saturating | you hand ready buyers to competitors |
| Create new demand | the largest slice, if the harvest ceiling is low | the harvest dries up in two months |
| Win back past visitors | 10 to 20% | you pay twice for the same lead |
| Test a new channel | 5 to 10% | you discover the next channel too late |
The last row is the one everyone cuts first, and the one that secures next year.
The mistake of reallocating too fast
Changing the split every week keeps every channel permanently in learning. Each large change costs days of unstable delivery, and if you change weekly, no channel ever gets to work well.
Monthly, with at least 30 days of data. And move 10 to 20% of the total at a time, not half.
Legitimate exception: a channel spending with zero sales for three straight weeks. That is not reallocation, that is correction.
The test nobody wants to run
The only honest way to know what a channel contributes is to switch it off and watch.
Two to four weeks with a channel paused show what happens to the total — not to that channel's number, which obviously goes to zero, but to the whole company's sales.
If you switch social off and total sales fall 30% while social accounted for 20% of attribution, it was worth more than attribution showed. If they fall 5%, it was worth less.
It is expensive, it is frightening, and it is the only experiment that genuinely answers. Do it with the smallest channel first, in a month without an aggressive target, and with the client informed.
To sum up
Do not split by CPA, split by ceiling and by function. Compare in the CRM, never in the dashboards. Give each channel enough budget to be judged, or switch it off. Reallocate monthly, not weekly, in slices of 10 to 20%. And reserve a small share to test the next channel, because the channel that is cheap today will not be cheap forever.
For the decision to raise spend within a channel you have already chosen, see scaling without breaking.
Frequently asked questions
Should I put more budget into the channel with the lowest CPA?
Not automatically. The cheapest channel may be harvesting demand that already existed, while the expensive one creates new demand. Both are needed, and the cheap one saturates first.
How do I compare channels when each has different attribution?
Compare in the CRM, not in each platform's dashboard. A sale recorded in the CRM with the right source is the only common basis across channels.
What is a starting split for someone just beginning?
Start with the channel where demand already exists, usually search, and only add a second channel once the first is stable and saturating.
How often should I move budget between channels?
Monthly, with at least 30 days of data. Moving weekly reacts to noise and keeps every channel permanently in learning.