Most marketing forecasts start their lives as a back-of-envelope ratio. Total bookings divided by total spend gives you cost-per-booking. Multiply next quarter's planned spend by that ratio, and you have a forecast. It's fast, intuitive, and wrong in ways that compound.
For an international travel platform, the ratio-based forecast was driving real decisions across more than a dozen markets. Should we shift budget from France to the UK? Increase SEM in Germany? Test Latin America? The model was making three assumptions that weren't true: that bookings would fall to zero without marketing, that all channels were equally productive, and that the relationship between spend and bookings was the same in every country.
Why a Ratio Breaks Down
The ratio approach assumes a linear pass-through: every additional dollar produces the same incremental return as the last. Most paid media exhibits diminishing returns instead. Most categories have a meaningful organic baseline. Different countries have wildly different responsiveness curves.
A ratio model implicitly assumes you're buying the average dollar. In reality, you're always buying the marginal dollar — and marginal economics are almost never the same as average economics.
The practical consequence: ratio-based forecasts systematically overstate the impact of small spend changes (because they ignore baseline) and understate the cost of large changes (because they ignore diminishing returns). For a global team trying to allocate a fixed budget across countries, those errors push spend toward exactly the wrong places.
The Updated Approach
We replaced the ratio with country-level regression models relating spend to bookings, controlling for seasonality, channel mix, and country-specific baseline demand. Each country gets its own response curve, with its own intercept (baseline bookings without marketing), its own slope (marginal lift per dollar), and its own diminishing-returns shape.
That last piece changes the planning conversation. Instead of "what's our cost-per-booking globally?", the team could ask "where on its curve is each country?" — and immediately see which markets had room to scale efficiently, which were already saturated, and which had baseline demand strong enough to coast on.
What the Forecasts Showed
Two findings landed early. First, implied ROAS varied dramatically across countries — far more than intuition had suggested. Markets that looked productive on a ratio basis were already past their efficient frontier; markets that looked weaker were further down their curve and had meaningful headroom. Shifting incremental dollars from saturated markets to under-served ones produced more total bookings at the same spend.
Second, channel mix within each country mattered as much as the country-level allocation. Some countries were over-indexed on SEM at a point where SEM had hit diminishing returns; others had under-invested in channels still productive. The same regression framework informed both decisions, with the same logic.
From Forecast to Always-On
The model's first deliverable was a quarterly forecast. The longer-term deliverable was something more useful: a continuously-updated forecasting layer integrated into the team's BI tooling, so ROAS estimates and bookings projections refreshed as new data flowed in. That changed the cadence of planning — from a once-a-quarter exercise based on a stale model to an ongoing read on which countries were pacing as expected and which were drifting.
It also created a foundation for testing. A forecasting model with proper baselines makes it possible to design experiments — incremental spend in a test market, holdouts, geo splits — and read the results against an actual counterfactual rather than a rough ratio. Over time, that's where the leverage compounds.
Forecasting marketing across regions or channels with a ratio that's stopped working? Say hello.