Marginal CAC is the cost of acquiring the next customer at a brand’s current spend level — the slope of the spend-vs-new-customers response curve at the point the brand is operating today. Blended CAC divides total acquisition spend by total new customers across every tier already deployed; marginal CAC reads the curve where the next dollar lands. The two numbers usually disagree, and the disagreement is the whole point of separating them.
Diminishing returns are why. The first dollars into a channel buy its easiest-to-convert audience — on Meta, that’s retargeting pools, lookalikes of best customers, warm interest segments. As spend climbs, the audience the platform has left to show ads to gets colder, and each incremental dollar produces fewer new customers. A Meta program running at $50k/week with $40 blended CAC may have $80 marginal CAC: the next $10k of spend brings in roughly 125 new customers (about $80 each), not 250 at the blended rate. Operators who use blended CAC to decide whether to push more budget into a saturated channel systematically over-invest, because the average drags the next-dollar cost downward in the analysis.
The scaling decision compares marginal CAC against marginal LTV — or, more practically, against payback period and contribution margin tolerances. If the next $10k of Meta spend produces customers at an 8-month payback when the brand’s cash position underwrites 4 months, that spend is wrong even if blended payback still looks healthy. The threshold depends on channel mix, gross margin, and growth-vs-cash posture; these are operator conventions, not laws. The constant is that the decision is made on the margin, not the average.
Measurement is noisier than the formula suggests. The cleanest read is an incrementality test that varies spend up or down and measures the new-customer delta. The practical approximation is spend-tier regression on historical data with channel and seasonality controls — the response-curve estimation that marketing-mix models do explicitly. Both are noisy at the daily level and smoother at the weekly level; pin reads to a window long enough to swamp run-to-run variance.
One caveat that’s easy to skip: marginal CAC is channel-specific. Every channel has its own response curve, its own saturation point, and its own audience overlap with the rest of the mix. A portfolio-wide marginal CAC averaged across channels is rarely useful for any decision a budget meeting actually makes. The channel-level reads are.