Three sources, three different stories
Ecosh Life is Estonia's first dietary supplement manufacturer. It sells mainly through its own online store and advertises on Google Ads and Meta Ads.
In the first half of 2024, Google Ads claimed 51% of Ecosh's revenue and Meta Ads claimed 54%. GA4, the supposed source of truth, credited Meta with just 5.89% and gave Klaviyo email 13.7%. The numbers couldn't all be right, and every budget decision was being made on data that contradicted itself.
GA4's last-click model gives the credit to the final step before a purchase: organic search, email, a direct visit. Taken at face value, it said to cut Meta. But Meta could well have been the channel building the brand interest that all the others were converting.
A Marketing Mix Model built for Ecosh
Instead of trusting any one platform's count, our data scientist Chinmay Kulkarni built a Bayesian Marketing Mix Model to estimate what each channel adds on its own. Gertrud Rei ran Google Ads, and our cooperation partner Andrus Kiisküla ran Meta Ads.
- →A causal discovery algorithm, checked against what we know about the business, to map how channels, brand interest and revenue affect each other
- →Ad stock, to capture what an ad keeps doing days and weeks after someone sees it
- →Saturation curves, to show where extra spend starts bringing in less
Meta's marginal return was 2.5× Google's
Once long-term and indirect effects were counted, Meta Ads had the strongest positive impact on revenue of any channel. Each extra euro on Meta brought in almost 2.5× as much additional revenue as an extra euro on Google Ads, in the mid-spend range Ecosh was working in.
GA4 couldn't see this because Meta worked indirectly. It built interest in the brand, and that interest turned into sales later through organic search, direct visits and email, the channels GA4 gave the credit to. The model also showed where returns started to flatten, so the team knew how far to push spend.
Testing budgets before spending them
With the model in place, we built a forecasting system that predicts next month's revenue from the planned spend and dates. It runs thousands of Monte Carlo simulations and gives a worst case, a most likely case and a best case. Each month the team looked at the risk and likely return before deciding where the budget went.
The strongest period in Ecosh's history
With the model guiding budget from September to December 2024, Ecosh grew revenue 56% on the same months a year earlier and passed its annual growth target in four months. That stretch included two of the highest revenue months the company has had, and the year closed with its highest annual revenue on record. ROAS rose 6%, while media spend fell from 13.6% to 12.8% of revenue. The project won Digitegu 2025.