THE MARGINAL BUDGET ALLOCATION CHEAT SHEET
Marginal budget allocation means funding every channel on the return of the next pound you spend. The average return of the pounds you have already spent is the wrong number. Most teams misallocate budget by backing the channel with the best average return. The right move is to fund the channel with the best marginal return at current spend.
The maths is unforgiving. A channel showing 200% average ROI at £150K a day can be losing money on every pound past £115K. Simply not spending that last £35K a day moves ROI from 180% to 239%. And splitting £150K across two search engines by equalising marginal return (£95K + £55K instead of £150K + £0) lifts ROI from 180% to 343%. That is the worked example above: reallocation alone nearly doubles ROI without spending an extra pound. I learned this way of thinking at eBay, from Mike Osborn, John Koryl and Tom Tang.
Not to Be Confused With
📊 Average ROI
Average ROI describes money already spent; marginal ROI decides where the next pound goes. A channel can show 200% average ROI while losing money on every pound past the margin. The average is a rear-view mirror, not a steering wheel.
🧪 Incrementality
Incrementality asks whether your spend worked at all. Marginal allocation asks where the next pound works hardest. You need the first answer before the second question means anything.
📅 Annual budget planning
This is not a once-a-year spreadsheet exercise. Curves move as markets, competitors and creative change. Reallocate continuously, in small steps, and re-measure after every shift.
Diminishing Returns Visuals (Use these in planning)
Response Curves by Channel
Marginal Return Curves (Differential)
The other half: lifetime value
Everything so far is cost-side thinking: what the next pound buys. The other half is what the customer is worth. A pound that acquires a customer worth £500 over their lifetime is a very different pound from one that acquires a £20 customer, so I compare channels on incremental profit per pound, with lifetime value sitting inside every marginal-return estimate.
How to allocate budget each cycle
- Estimate current marginal return by channel using experiments, lift studies, and calibrated MMM.
- Rank channels by next-pound return, not blended CAC/ROAS.
- Move spend in small steps (e.g., 5–15%) to avoid overshooting the optimum.
- Re-measure after each shift because curves move as market and creative change.
- Lock guardrails: minimum brand presence, geo caps, and saturation ceilings.
Do this vs avoid this
✅ Do this
- Use incremental profit/revenue per extra pound spent.
- Reallocate continuously, not annually.
- Separate steady-state spend from test spend.
- Pre-register read windows before calling winners.
❌ Avoid this
- Funding channels with best historical average ROAS.
- Reading attribution as causal truth.
- Making large one-shot shifts without retesting.
- Ignoring diminishing returns in planning decks.
Quick decision table
| Observed condition | Action |
|---|---|
| Channel X marginal return > Channel Y by meaningful margin | ✅ Shift incremental budget from Y to X |
| Channel X high average ROAS but low marginal return | ❌ Do not scale X further; harvest and hold |
| All channels near flat marginal return | ✅ Invest in creative/product improvements before raising spend |
| Evidence quality weak / conflicting | 🧪 Run bounded lift test before reallocation |
Know Its Limits
Marginal curves move. Creative fatigues, competitors bid up your keywords, seasonality shifts demand: a curve you estimated last quarter is a rumour, not a fact. Re-measure continuously, and treat every allocation as provisional.
Protect a learning budget the optimiser cannot touch. If every pound must clear a marginal-return hurdle, no new channel ever gets its first pound. Left alone, the portfolio slowly converges on yesterday’s winners.
My position: this framework hill-climbs. It will find the best possible allocation across the channels you already run, and it will never tell you to invent a new one. I have seen teams perfectly optimise a portfolio that was slowly dying: every marginal decision was correct, and the strategic question never got asked.