Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Digital Marketing › Marketing Strategy

Budget Allocation Across Channels

How to split a marketing budget across channels and months by marginal return, with reserves for testing and demand creation and rules for moving money.
Guideintermediate8 min readUpdated

Why averages mislead

A report shows each channel's return on ad spend. The obvious move is to shift money toward the best one. It is usually wrong, for a reason the average hides: returns diminish. The first spend in brand search reaches people already looking for you, at a very low cost per acquisition; the last spend in brand search is bidding against yourself for the same people. Paid social reaches the most responsive audience first and progressively less responsive ones after. Every channel has a curve, and the average sits somewhere on it, saying nothing about the slope at the current spend.

  1. Spend increases
  2. Audience saturates and auction prices rise
  3. Each extra unit returns less
  4. Marginal return falls below the next-best channel's
  5. Move the next unit there
Allocation stops where marginal returns are equal across channels.

The practical rule: allocate so that the marginal return — the CPA or ROAS of the next increment — is roughly equal across channels, not the average. Where the marginal figure is unknown, the curve has to be estimated or tested.

Estimating the curve

Three sources, in order of reliability:

  1. Spend-and-response tests. Raise spend in a channel by a known step for a period and measure the change in outcomes; repeat at another level. Platform experiment tools and geographic splits do this with controls. The result is a few points on the curve, which is enough.
  2. Historical variation. Past periods with different spend levels in the same channel, adjusted for season and for what else was running. Noisier, but free.
  3. Platform forecasts and planning tools. Reach and frequency planners, budget simulators and auction insights indicate where saturation begins for a given audience. Treat them as the platform's estimate, not as fact.

The CPA calculator and ROAS calculator help turn these into the figures the comparison needs; the break-even calculator gives the floor below which a marginal return is a loss regardless of how it compares with other channels.

A marketing mix model is a further source for businesses with the history to fit one, and the guide to evaluating a marketing mix model sets out what to check before its curves are trusted in an allocation.

The allocation method

  1. Start from the mix. Each channel has a job from the channel mix; allocation funds the jobs, it does not re-decide them.
  2. Set the floors. The minimum each channel needs to do its job at all: enough for the algorithms to learn, enough frequency to be noticed, enough volume to measure. Below the floor, spend is wasted; a channel that cannot be funded to its floor should be cut rather than starved.
  3. Set the reserves. A testing reserve (new channels, new audiences, new creative, held separately from channel budgets) and, where the business depends on future demand, a demand-creation share protected from short-term return comparisons — the brand vs performance comparison covers why.
  4. Allocate the rest by marginal return. Fund the channel with the best marginal figure until its marginal figure falls to the next channel's, then fund that one, and so on.
  5. Check against the target. The budget calculator shows what the allocation implies in outcomes at the assumed rates; if it does not reach the objective, the objective, the total or the rates are wrong, and the plan should say which.
  6. Phase over time. Spend follows demand: more in peaks, less in troughs, with the peaks funded early because auction prices rise with everyone else's spend. The campaign calendar is where the phasing becomes dates.
Illustrative example — not a real result

Reallocating an e-commerce budget

A retailer's report shows brand search with the best average return, paid social the worst, and non-brand search in between. A spend test shows brand search is saturated — the last step up in budget produced almost no extra orders, only higher cost per click — while paid social, at its current low spend, is still reaching its most responsive audience and the step-up test produced orders at a marginal cost better than non-brand search's. The new allocation holds brand search at its floor, moves the freed budget to paid social up to the point where its marginal cost matches non-brand search, and keeps the testing reserve intact. Average returns by channel now look worse; total orders for the same budget are higher. Reported this way, the change is defensible; reported as "we cut the best channel", it is not.

A worked allocation

Illustrative example — not a real result

Allocating a quarter's budget across three channels

A retailer has a fixed quarterly budget and three channels with jobs: brand search (capture), non-brand search (capture) and paid social (create). Spend tests over the previous two quarters give a few points on each channel's curve. Brand search converts cheaply at its current spend, but the step-up test showed almost no additional orders — it is at its floor and saturated, so it is held there. Non-brand search's marginal cost per order rises steadily with spend; paid social's marginal cost is higher than non-brand search's at low spend but falls as the algorithm learns and then rises past a certain level. The allocation funds brand search to its floor, funds paid social up to the point where its marginal cost per order equals non-brand search's, puts the remainder into non-brand search, and holds the testing reserve and a protected creation share separately. Phasing puts more into social in the first month (to build the audience that search will capture in the peak) and more into search during the peak. The plan records the marginal figures it assumed, so the quarterly review can compare them with what happened.

Allocation review checklist

0/8

Allocating with no history

A new programme has no curves to estimate, and the temptation is to split evenly. Better: fund the channel closest to existing demand (usually search) to its floor first, because its return is the most predictable; put a deliberately small amount into one creation channel as a test rather than a commitment; and keep the reserve larger than usual, since the first quarter's job is to produce the points on the curve that the second quarter will allocate on. Treat every initial figure as a hypothesis with a review date, and resist adding channels until the first ones have produced a measurable marginal return. The guide to marketing decisions before you have data covers the rest of that first period, from targets set from the margin to the point at which estimates become a baseline.

Reallocation rules

Money should move on evidence, under rules written before the results arrive, so that reallocation is a process rather than an argument. A workable set:

  • Rhythm: review monthly; weekly in peak periods; never daily.
  • Thresholds: move budget only when a channel's marginal figure has been above or below the agreed band for the full review period, not on a single week.
  • Step size: move a fixed share at a time, large enough to matter and small enough to reverse.
  • Floors and reserves are not raided. A channel below its floor is cut, not trimmed; the testing reserve is spent on tests or returned, not absorbed.
  • Attribution caveat: judge capture channels on attributed return; judge creation channels on reach and on the demand trend, because last-click attribution under-credits them. Where it matters, Incrementality tests settle the argument.

When a move is agreed, write it down: the budget reallocation memo template records the amount, the marginal evidence, the constraints, the reversal thresholds and the approval, so the next review can judge the move rather than remember it.

Allocation across time, not only channels

Two temporal allocations are as important as the channel split:

  • Seasonality. Demand is not flat; neither should spend be. Historical demand by month (search volume, sales, site traffic) gives the shape; the budget follows it, with awareness activity leading the peak and capture activity concentrated in it.
  • Front-loading versus pacing. Launches and tests want spend concentrated early to learn fast; always-on activity wants even pacing to keep algorithms stable and frequency sensible. Mixing the two in one budget line hides which is which.

A business with branches, franchisees or dealers often sets part of the budget aside for each site to spend locally, and the guide to running a shared local marketing fund covers the rules, claims and reporting that keep that money from going unspent or unmeasured.

Organic search is the clearest case of costs arriving long before returns, so it rarely wins a single-quarter allocation; the guide to building an SEO business case shows how to stage that commitment with review points and compare it with other channels over a fair horizon.

To size the money that belongs in the busiest weeks, the marketing budget calculator for seasonal peaks uses last peak's acquisition cost and a short window.

When budget moves after every good or bad week and the quarter's plan no longer holds, the diagnosis of a marketing plan that keeps changing shows where reallocation rules fit among the other causes.

Where an allocation that added budget to a channel produced no extra sales, the page on increased ad spend with flat sales shows how to read the marginal return of the added money before moving it again.

Common mistakes

  • Allocating on average return. Over-funds saturated channels, under-funds growing ones.
  • No floors. Channels funded below the level at which they can work, then judged as failures.
  • Raiding the reserves. Testing stops, demand creation stops, and next year's capture channels get more expensive.
  • Flat monthly budgets. Starving peaks and overspending troughs.
  • Daily reallocation. Chasing noise and never letting the platforms learn.
  • One blended metric. A single ROAS across channels with different jobs hides everything useful.

For how the total is set before allocation, see marketing budgeting methods; for measuring the programme's return as a whole, see the marketing ROI page.

Frequently asked questions

Should the budget go to the channel with the best ROAS?

Only up to the point where its return starts falling. The channel with the best average return is usually the one closest to saturation; the next unit of spend there often returns less than the next unit in a channel with a worse average. Allocation is about the margin, not the average.

How much should be reserved for testing?

A share large enough to learn something and small enough that losing it entirely does not threaten the target. The common practice is a modest fixed share of the total, agreed in advance and protected from being raided when results are tight.

How often should budget move between channels?

On a fixed rhythm — monthly for most programmes, weekly inside peak periods — under rules written in advance. Moving money on every daily fluctuation chases noise; never moving it ignores evidence.

Terms used on this page

Tools to try

Frameworks

Where this page is used

Part of these learning paths

Also related

Related