Start for freeSign in

GlossaryAds

The return on the next baht of ad spend, not the average across all of it. The number that decides whether to scale a campaign.

Marginal ROAS is the additional revenue produced by the next baht of ad spend, as distinct from average ROAS, which divides all revenue by all spend. Because advertising saturates — the cheapest, highest-intent impressions are bought first and each additional baht reaches a progressively less interested audience — marginal ROAS is always lower than average ROAS on a campaign that is spending anything at all. Every scaling decision is a marginal question: not “is this campaign profitable?” but “would one more baht in it be?”. A campaign can have an excellent average ROAS and a marginal ROAS already below break-even, which is the precise condition under which increasing its budget destroys money.

Formula

Marginal ROAS = d(revenue) / d(spend)

Saturating response, Hill form:
  orders(s) = ceiling × s / (k + s)
  k = the spend at which response reaches half its ceiling

Optimal spend is where marginal profit = 0, i.e.
  marginal ROAS × contribution rate = 1

Example

A Shopee campaign at ฿10,000/day returns ฿60,000 — average ROAS 6.0.
Raise it to ฿12,000/day and revenue moves to ฿66,000.

Marginal ROAS = (66,000 − 60,000) / (12,000 − 10,000) = 3.0
At a 20% contribution rate, break-even is 5.0.
The average says scale. The margin says the last ฿2,000 lost ฿1,800.

Why does marginal ROAS fall as you spend more?

Because impressions are not interchangeable. An auction serves the query where your product is the obvious answer before it serves the query where it is a plausible one, and the buyer who searched your exact model number converts at a different rate from the buyer who searched the category. As budget rises the campaign reaches further down that ordering, so each additional baht buys a less valuable click. Modellers capture this with a saturating response curve — the Hill or Michaelis–Menten shape, orders proportional to spend divided by spend plus a half-saturation constant — which rises steeply from zero and flattens toward a ceiling. The important property is not the exact algebra but its consequence: there is always a spend level past which the next baht loses money, and average ROAS gives no hint of where it is.

How do you find the profit-maximising spend?

The optimum is where marginal profit crosses zero, which happens when marginal ROAS multiplied by the contribution rate equals one. Above that point the next baht returns less contribution than it costs. Two things follow that surprise people. First, the profit-maximising spend is almost never the ROAS-maximising spend: ROAS is maximised near zero spend, where only the best impressions are bought. Second, when several campaigns share a budget, profit is maximised when their marginal returns are equal, not when their average ROAS figures are equal — the equimarginal principle. A portfolio balanced on average ROAS is systematically mis-allocated.

stop scaling when:  marginal ROAS = 1 / contribution rate

  contribution 20%  →  stop at marginal ROAS 5.0
  contribution 40%  →  stop at marginal ROAS 2.5

across campaigns: equalise marginal return, not average ROAS

Can you measure marginal ROAS without an experiment?

Partially, and the limits matter. Fitting a saturation curve to observed spend and orders gives a usable slope near the range of spend the campaign has actually operated in. Outside that range it is extrapolation, and the ceiling parameter in particular is only identified if the campaign has been pushed near saturation at some point — a campaign that has never spent past the steep part of its curve simply does not contain the information needed to say where its ceiling is. Honest practice is to trust the marginal estimate close to current spend, treat large jumps as uncertain, and use deliberate spend variation, not just historical drift, to learn the shape.

Why it matters

Almost every budget conversation in marketplace advertising is conducted in average ROAS, and average ROAS cannot answer the question being asked. “This campaign is at ROAS 6, let us double it” is a marginal claim supported by an average number. The gap between the two is where scaling goes wrong: the average stays respectable for a long time after the marginal return has gone underwater, because it is still being propped up by the cheap conversions bought at the start.

In DataGlass

DataGlass fits a saturating spend-response curve per Shopee campaign and allocates budget by marginal return rather than by average ROAS, so a campaign that is already flat gets held while one still climbing gets funded. The response curve and the implied stopping point are shown as evidence on the recommendation, along with how far the estimate is extrapolating beyond observed spend.

Where this shows up

Related terms

  1. [1]Diminishing returns — overviewThe general economic property that makes marginal return fall below average return whenever a factor of production is increased against fixed complements.
  2. [2]Equimarginal principleThe allocation rule behind cross-campaign budgeting: total return is maximised when marginal returns are equalised across uses, not when averages are.
  3. [3]Jin, Wang, Sun, Chan & Koehler — Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects (Google Research, 2017)Uses the Hill function to model advertising shape effects and saturation, and discusses the identification difficulty when observed spend does not span the curve.

See it computed from your orders, fees and ads, with a demo on your data.

Book a demoSee pricing