Marginal return: place the next euro, do not reward the past.
Answer:
Average return divides cumulative outcome by cumulative spend. Marginal return measures the outcome from a small additional spend at the current level. Allocation should bring risk-adjusted marginal returns closer under constraints, rather than move budget to the historically best ROAS.
Method and decision:
Estimate spend-to-contribution curves with lag, saturation and context effects, publishing observed range and uncertainty. Compare R(s + Δ) − R(s) for a buying-compatible increment. Exclude moves violating minimums, maximums, available reach, contracts, creative quality or commercial capacity. Re-estimate after every meaningful increment because allocation changes the slope point. Move budget gradually from low marginal-return zones to higher ones until equalisation, a constraint or excessive uncertainty.
Worked example:
At current spend, channel A yields €0.60 per additional euro and B yields €1.40. A €50,000 tranche can leave A without crossing its minimum. Projected A loss is 50,000 × 0.60 = €30,000. B gain is 50,000 × 1.40 = €70,000. Net contribution gain is €40,000 before transition cost and uncertainty interval. Make a first step only if the prudent bound remains positive, then recalculate both slopes; never extrapolate €1.40 to every future tranche.
Limits:
Curves estimated from little variation can have unstable slopes. Channel interactions make separate optimisation incomplete. Contracts, creative work and buying capacity can make a mathematical optimum infeasible.