Optimising Marketing Spend With AI: Allocate by Marginal Return

Average ROAS answers the wrong question

Return on ad spend (ROAS) is attributed revenue divided by ad spend. Reports show it as an average across everything spent in a channel. But budget decisions are about the next pound, and the next pound in a channel almost never earns the average: the cheapest, most interested buyers are reached first, and each extra increment reaches people who are harder to convert.

So the question for allocation is: what did the last increment of spend in each channel return, and is that above or below the level at which you make money?

Step 1: Know your break-even ROAS

Break-even ROAS is the ROAS at which gross profit from the revenue exactly covers the ad spend: 1 ÷ gross margin.

• • At a 40% gross margin, break-even ROAS is 1 ÷ 0.40 = 2.5.
• • At a 60% margin it is about 1.67; at 25% it is 4.0.

Any spend returning less than break-even loses money on first purchase, however good the average looks. Use the calculator to check a channel's ROAS, break-even ROAS and contribution after ad spend.

Calculator

ROAS and break-even calculator

Check whether ad revenue is profitable once your gross margin is taken into account.

Ad spend (£)Revenue attributed to the ads (£)Gross margin (%)

The starting figures are illustrative round numbers, not AfM prices or client results. Replace them with yours.

ROAS

3.00x

Attributed revenue divided by ad spend

Break-even ROAS

2.50x

1 divided by gross margin: below this, the ads lose money before any other cost

Contribution after ad spend

£1,000

Revenue × margin, minus ad spend

Assumptions

• Attributed revenue depends on the attribution model; platform-reported revenue often overlaps between channels.
• Contribution excludes fees, returns, fulfilment you have not counted in margin, and overheads.

Step 2: Estimate the marginal return in each channel

Ways to estimate what the last increment earned, from quickest to most reliable:

1. • Compare budget steps. When a channel's budget went up or down, compare the change in attributed revenue with the change in spend over similar weeks.
2. • Platform simulators. Some ad platforms estimate results at different budget levels; treat them as indications.
3. • Lift tests. Meta's Conversion Lift, for example, compares people shown ads with a holdout who were not, measuring incremental conversions rather than attributed ones.
4. • Media mix modelling. Google's open-source Meridian model uses Bayesian methods to estimate each channel's incremental contribution and response curves from your historical data, and offers budget scenarios. It needs a reasonable history of spend and outcomes to be useful.

Worked example

Illustrative example with hypothetical round figures. A retailer with a 40% gross margin (break-even ROAS 2.5) runs two channels.

• • Channel A: £10,000 a month, £40,000 attributed revenue, average ROAS 4.0. The last £2,000 added this quarter brought £4,000 of extra revenue, a marginal ROAS of 2.0.
• • Channel B: £5,000 a month, £15,000 attributed revenue, average ROAS 3.0. The last £1,000 added brought £3,500, a marginal ROAS of 3.5.

On averages, A looks like the better channel. On margins, A's last £2,000 is below break-even:

• • Contribution from A's last £2,000: £4,000 × 40% minus £2,000 = minus £400.
• • A overall is still profitable: £40,000 × 40% minus £10,000 = £6,000. Enter £10,000, £40,000 and 40% in the calculator above and it shows ROAS 4.00x, break-even 2.50x and £6,000 contribution; enter the last increment alone (£2,000, £4,000, 40%) and it shows 2.00x and minus £400.

Reallocation. Move £2,000 from A to B. Assume B's next £2,000 returns a little less than its last increment because of diminishing returns, say a marginal ROAS of 3.0.

• • Revenue lost from A: £4,000
• • Revenue gained in B: £2,000 × 3.0 = £6,000
• • Net revenue change: plus £2,000 at unchanged total spend
• • Contribution change: £2,000 × 40% = plus £800 a month

Average versus marginal view (illustrative figures)

Measure	Channel A	Channel B
Monthly spend	£10,000	£5,000
Average ROAS	4.0	3.0
Marginal ROAS on last increment	2.0	3.5
Break-even ROAS at 40% margin	2.5	2.5
Verdict on the next pound	Below break-even: reduce	Above break-even: increase and re-measure

Move money in steps and re-measure after each. B's marginal return will fall as it grows; stop when the two channels' marginal returns are roughly equal or B's falls to break-even.

A trap: high ROAS on demand you already had

Some channels show very high ROAS because they capture people who were going to buy anyway. Brand search is the usual example. The Mumshandmade case record, work Jakub delivered before founding AfM, lists Google Brand Search ROAS of "34.17X" since the campaign's December 2024 launch, as reported by Google, and explains that brand search captured people already searching for the brand after seeing its Meta ads. The record itself attributes the demand to earlier Meta exposure. That is the right way to read any brand search figure: it is largely harvesting demand created elsewhere, so cutting the channel that created the demand would eventually shrink the brand search result too. Marginal and incremental thinking protects you from starving the source to feed the harvester.

What AI decides, and what it does not

• • Within a channel, platform bidding systems already use machine learning to set bids towards the conversion signal you give them. The most valuable thing you can do is give them the right signal: a conversion that reflects profit or qualified pipeline, not a cheap proxy.
• • Across channels, the platforms each report their own attributed conversions, and each has an interest in its own share. Allocation between them is your decision, informed by marginal estimates, lift tests and, at enough scale, a media mix model.
• • AI-assisted analysis can speed up the work: pulling spend and revenue by week, flagging diminishing returns, drafting scenarios. A person still decides and owns the outcome.

For why platform-attributed conversions cannot simply be added up, see AI Marketing Revenue Attribution. For cutting production costs rather than moving media, see Reduce Marketing Costs With AI.

FAQ

Should I ever run a channel below break-even ROAS?

Sometimes, if repeat purchases or lifetime value make up the difference and you have evidence for it. Then use a break-even based on the profit you can prove over a set period, not first purchase alone.

How big should each reallocation step be?

Small enough that you can read the effect, often 10% to 20% of the channel's budget, and held long enough for the platform to settle before the next step.

Is a media mix model worth it for a small business?

Usually not at first, because it needs a reasonable history of spend and outcomes across channels. Budget steps and lift tests give most of the benefit at lower cost.

Does Smart Bidding make manual allocation unnecessary?

No. Bidding algorithms optimise within a campaign towards your chosen conversion. They do not decide how much each channel should receive relative to the others.