One of our customers was losing over €167,000 per month on Meta without noticing it.

Their blended Meta ROAS looked fine and within their target so they kept scaling.

The loss only showed up once you started diving deeper into the data to understand how they allocated their money and modeled saturation curves per country to look at what each next euro returns.

This issue is the breakdown. What saturation curves are, why blended metrics hide this much money, and how to validate big budget shifts before you make them:

Many brands set their Meta budget on a total level, then split it by country based on revenue share.

Sounds reasonable at first. But most businesses are actually multiple smaller businesses at the same time. Each market has its own inputs and outputs. Its own competition, its own CAC, its own ceiling.

One of 9-figure customer runs Meta across 8 countries. So the first thing we did was create 8 separate views in Klar, one per market, to track each smaller business unit on its own.

That alone showed 43% differences in Meta ROAS between markets.

Now you could just start cutting spend in countries with a lower ROAS and move it to countries with a higher ROAS. And that would move you in the right direction.

But the next question that follows from this is:
How much should I cut in one country and how much should I increase it another.

To build an understanding for that, we went one level deeper and built channel saturation curves per country.

A saturation curve answers one question: How much will the next euro of spend return? Not the average euro but the next one.

The differences were even larger than the ROAS gaps. Marginal ROAS ranged from 1.69 in the worst country to 3.94 in the best.

Same channel. Same brand. Huge difference in what the next euro returns.

You want every country operating at the same marginal ROAS.

That's when your budget is fully optimised. You shift from countries with a lower marginal ROAS to the ones with a higher marginal ROAS until they all reach the same point.

For some countries, that meant increasing spend by 247%, while for others cutting spend by over 50%.

Simply shifting budget between countries would increase total channel ROAS by roughly 13%. Same total spend, no new creative, no new channel.

(We built safeguards into the optimization that cap the maximum change per country based on available historic performance)

Playing through these optimisation scenarios is the biggest strength of Marketing Mix Modelling (MMM).

But an MMM is still a model.

And before you move millions based on a model, you validate.

To pressure-test the model, we are now running geotests (more on that in this issue) in three markets:

Because, as these saturation curves have shown, those curves and therefore incrementality are oftentimes very different from one market to another.

If all three confirm the recommendation, those big shifts can be made with a much higher confidence level.

Making big changes is key to getting big results.

But mitigating that risk beforehand through incrementality tests limits the risk and therefore the downside of such a change, should it be wrong.