Key takeaways
- Average ROI describes historical efficiency; marginal ROI estimates the return near the next unit of spend.
- Response curves carry assumptions about cost, flighting, lag, and the range supported by historical data.
- Optimization needs realistic floors, ceilings, commitments, capacity, and scenario uncertainty.
- Make reversible budget moves and compare outcomes with the assumptions that justified them.
01
Do not drive forward while staring at an average
A channel can show strong historical ROI and still be close to saturation. Another can show a lower average yet offer better return on a modest additional investment. Meridian defines ROI as historical incremental outcome divided by spend and marginal ROI as the change in incremental outcome around an additional unit of spend.
The distinction matters because budget allocation is a forward decision. Ranking last quarter’s channel averages is not the same as estimating what happens near the proposed spend level next quarter.
Swipe to compare every column
| Measure | Useful question | Common misuse |
|---|---|---|
| ROI | How efficient was historical spend? | Assuming the next dollar performs at the average |
| Marginal ROI | What return is estimated near current spend? | Treating a local estimate as valid at any budget |
| Response curve | How might outcome change across spend levels? | Ignoring uncertainty and unsupported extrapolation |
| Scenario | What follows under stated future assumptions? | Presenting a conditional plan as a forecast guarantee |
02
Read the assumptions printed beneath the curve
Meridian notes that response curves scale historical media units while preserving the observed flighting pattern, and that its marginal-return definition assumes a constant cost per media unit at the historical average. Future auction prices, creative quality, inventory, targeting, seasonality, and execution may differ.
Show the historical spend point, modeled support, credible interval, and proposed point together. A smooth line far beyond observed spending can create confidence the data never supplied.
03
Make the optimizer respect the business
Add contractual minimums, learning budgets, geographic coverage, audience saturation, production capacity, sales capacity, inventory limits, brand commitments, and maximum feasible changes. An unconstrained mathematical optimum may be operationally impossible or strategically brittle.
Run a fixed-budget case, a conservative case, and a downside case. Record the assumptions that change between them. The value of scenario planning is not a single perfect allocation; it is seeing which moves remain reasonable across plausible conditions.
04
Move in increments that can teach you something
Stage reallocations where possible. Preserve a baseline, annotate creative and targeting changes, and watch qualified outcomes rather than only platform conversions. Large simultaneous changes make it difficult to distinguish a wrong response curve from a different execution.
Return to the model with experiment evidence and realized outcomes. A recommendation that cannot be challenged by later evidence is not optimization; it is a decorative allocation spreadsheet.
Primary sources and further reading
Use the source material to validate details against your own context and current platform configuration.
- Google Meridian: Incremental outcome, ROI, marginal ROI, and response curves
- Google Meridian: Scenario planning and future budget optimization
- Google Meridian: Interpret optimizations
This field note follows the XenGrowth editorial policy: primary sources where available, visible limitations, material review dates, and no invented first-hand experience.
Stay with the problem
