What is budget scaling?

Budget scaling is the practical question a measurement programme exists to answer: given what we know, how much should we spend where? It is distinct from measuring incrementality, which establishes whether a channel works at all. Scaling asks what happens if you move the number — and the answer is governed by the shape of the response curve rather than by the average return you have historically achieved.

The critical distinction is marginal versus average. Diminishing returns mean each additional pound reaches a less responsive audience than the last, so the return on the next pound is always below the average across all pounds spent. A channel averaging 4.0 iROAS can have a marginal return under 1.0 at its current level, which means it is profitable overall and unprofitable to grow. Scaling decisions made on the average systematically overspend, and the error compounds because the average stays healthy while the marginal return deteriorates.

The optimal allocation follows a simple rule that is hard to execute: move spend until the marginal return is equal across channels. If one channel returns £3 on the next pound and another returns £5, moving money from the first to the second increases total return at no extra cost. Equilibrium is reached when no such move helps — and note this says nothing about the averages, which can differ substantially at the optimum.

Estimating marginal response is genuinely difficult and this is where most programmes stop short. Marketing mix modelling fits a curve to historical variation, which works when spend has varied enough to identify the shape and fails when it has been flat — a channel held at the same budget for two years contains almost no information about what a different budget would do. A scale-up test measures the marginal response directly and needs a large deliberate swing to detect anything, because the effect being measured is small by construction.

The practical sequence that works is to establish incrementality first, then marginal response, then reallocate. Scaling a channel whose baseline contribution has never been measured risks growing something that was never returning anything. And any scaling decision has a shelf life — response curves shift as audiences saturate, competitors change spend and creative fatigues — so the estimates need refreshing rather than treating as settled.

An advertiser has £2.4m across three channels and gross margin of 40%, making break-even marginal iROAS 2.5. Response curves are fitted from an MMM calibrated against geo tests, and the marginal return at current spend is computed for each.

Search averages 5.2 and returns 1.9 on the margin, which is the classic pattern for a saturated lower-funnel channel — the early spend captures high-intent demand cheaply and the later spend reaches people who were unlikely to convert anyway. At 1.9 against a 2.5 break-even, the last portion of search spend is destroying value while the channel as a whole looks like the star performer. Social is the opposite: a modest average and a marginal return of 3.4, meaning there is headroom. Moving money until the marginals converge is what the allocation rule prescribes, and £250,000 is the amount that brings them close without overshooting. Two cautions. These marginal figures come from a fitted curve, so their uncertainty is larger than the point estimates suggest, and a scale-up test on social would confirm the 3.4 before committing. And the curves shift — this allocation is right for now, not permanently.