Budget allocation by marginal return: a CFO-friendly guide

Most marketing budgets are built the same way: take last year's split, adjust each line by 5 to 15%, and defend it in a meeting. The result reflects habit and negotiating skill more than return. There's a better question to ask of every dollar: where does the next one earn the most? That's allocation by marginal return, and it's simpler than it sounds.

Average return vs marginal return

Suppose paid search returns $4 of incremental revenue for every $1 on average, and paid social returns $2.50. The tempting conclusion is "move money to search". But the average hides the shape of the curve.

Almost every channel shows diminishing returns. The first dollars reach the most responsive people: those searching for exactly what you sell, or the best-matched audiences. As spend rises, each extra dollar reaches people who are less likely to buy, at higher cost. So a channel with a high average return can have a low return on the next dollar, and a channel with a modest average can still have plenty of headroom.

The rule that follows is an old one from economics: keep moving money until the marginal return is roughly equal across channels. At that point, no reallocation can improve the total. Any budget where one channel's next dollar earns $3 and another's earns $1 is leaving money on the table.

A worked example

A company spends $100,000 a month across two channels. Through testing, it has estimated the incremental revenue each channel produces at different spend levels (illustrative numbers):

Today the split is $70,000 search and $30,000 social: $240,000 + $90,000 = $330,000 incremental revenue. Search has the higher average return (about 3.4 versus 3.0), so it looks like the winner. But its last $10,000 earns only $1.50 per dollar, while social's next $10,000 would earn $2.50.

Move to $50,000 and $50,000: $205,000 + $139,000 = $344,000. Same budget, about 4% more incremental revenue, and now the marginal returns are close ($2.50 and $2.40). That's the whole idea.

How to estimate marginal returns without a data science team

You don't need a perfect curve. You need a few credible points per major channel and a willingness to update them. Four practical sources, from most to least reliable:

1. Scale-up and scale-down experiments. Increase spend by 50 to 100% in a set of test regions, or cut it in others, and measure the incremental change with a geo test. Two or three tests at different spend levels sketch the curve. This is the gold standard because it measures causation, not correlation.
2. Platform lift studies at different budgets. Useful for single-platform curves, with the caveat that the platform measures itself.
3. Platform forecasting tools. Google's Performance Planner (opens in a new tab) projects how conversions change with spend, based on auction data. It estimates attributed conversions, not incremental ones, so apply a calibration factor from your tests (see iROAS vs ROAS).
4. Your own history. Weeks where spend changed for reasons unrelated to demand (a budget freeze, a launch delay) can hint at the curve's shape. Treat them as clues, not proof: spend often rises precisely when demand is high, which flatters the numbers.

Combine them pragmatically. Many mid-sized companies get most of the value from one well-designed scale test per major channel per year, plus calibrated platform forecasts in between.

A five-step allocation process

1. Define the objective in finance terms. Incremental contribution margin (revenue minus product, fulfilment and payment costs) is better than revenue. New customers may matter more than total orders. Agree on it with finance first.
2. Set a break-even line. With a 40% contribution margin, a dollar of spend must generate $2.50 of incremental revenue to break even on the first purchase. Adjust for repeat purchase value if you have reliable data.
3. Estimate marginal return at current spend for each channel, using the sources above, with a rough confidence level (high, medium, low).
4. Reallocate in steps. Move 10 to 20% of budget from the channels with the lowest marginal return to those with the highest, prioritizing high-confidence estimates. Never move everything at once: the curves are estimates.
5. Re-measure and repeat monthly or quarterly, and test any channel where a big move rests on a low-confidence estimate.

Ring-fence a test budget

Pure marginal-return allocation has a blind spot: channels you've never tried have no curve at all, so they never get budget. Fix it by setting aside a fixed share, often 5 to 10%, for structured tests of new channels, audiences or creative approaches, with clear success criteria. The test budget is how tomorrow's curves get drawn.

Real-world constraints

• Minimum effective spend. Some channels (connected TV, sponsorships, out-of-home) don't work below a threshold. Treat them as on/off decisions.
• Lagged effects. Upper-funnel spend can take weeks to show up. Measure over a long enough window, or the curve will undervalue it.
• Brand protection. Some branded search spend defends against competitors bidding on your name. Measure that specifically rather than assuming it.
• Seasonality. Curves shift in peak seasons, when demand is higher and auctions are more expensive. Plan seasonal allocations separately.
• Contracts and commitments. Upfront media buys and retainers limit how much you can move mid-year.

How to present it to a CFO

Finance leaders tend to welcome this approach because it speaks their language: incremental return on investment, at the margin, with stated confidence. A one-page summary works best:

• A table: channel, current spend, estimated marginal return at current spend, confidence, evidence (test name and date), proposed change.
• The expected total effect of the proposed reallocation, with a range.
• The tests planned to firm up the lowest-confidence estimates, and when results will be in.
• A clear rule for next time: "We reallocate quarterly based on marginal return, and we re-test any channel above X% of budget at least once a year."

What you're really offering is a process that gets more accurate every quarter, instead of a number defended once a year.

Where AI agents help

Agents are well suited to the unglamorous parts of this process: monitoring pacing against the plan, pulling spend and results into one view each week, flagging when a channel's efficiency drifts, and drafting the monthly reallocation memo for a strategist to review. The judgement calls (objectives, confidence levels and how much to move) stay human. See how to give AI agents a job in your marketing team.

Common mistakes

• Allocating by platform-reported ROAS. It rewards channels that capture existing demand. Use incremental estimates.
• Allocating by averages. The average return tells you about past dollars, not the next one.
• Making one giant move. Step changes let you check whether the curve behaves as expected.
• Never testing the big channels. The largest line in the budget usually has the most to gain from a test, and the most political protection.
• Forgetting margins. A revenue-maximizing allocation can be a profit-minimizing one.

Checklist

• Agree the objective (incremental contribution margin or new customers) with finance.
• Calculate your break-even incremental return.
• Estimate marginal return and confidence for every channel above 10% of budget.
• Reallocate 10 to 20% from lowest to highest marginal return.
• Ring-fence 5 to 10% for structured tests.
• Schedule a scale test for the largest low-confidence channel.
• Repeat quarterly.

Frequently asked questions

Need credible marginal-return estimates? Our incrementality testing service designs the tests and turns the results into a budget plan. Or talk to us about your next planning cycle.