What is Budget allocation?

A A$240,000 quarterly budget split 50 / 30 / 20 across three channels gives 240,000 × 0.50 = A$120,000, 240,000 × 0.30 = A$72,000 and 240,000 × 0.20 = A$48,000. The arithmetic is the easy half. The shares are the decision, and a split inherited from last quarter is a decision somebody made once and nobody has made since.

Budget allocation divides a total across channels, campaigns, markets and periods.

Channel budget = total budget × channel share

Nothing about that is difficult. The whole of the problem is in choosing the shares, and the default — last quarter's split, adjusted slightly — is a decision that was made once and has not been made since.

An account that allocates strictly on reported returns will drift in a predictable direction, and understanding why is most of what makes allocation a skill rather than a sort.

Attribution credits advertising that was present when a conversion happened. Branded search, retargeting and anything else that reaches people already close to buying are present by construction, so they collect credit efficiently and report the best numbers in the account.

The activity that created the demand those channels capture is furthest from the conversion, frequently outside the attribution window, and often produces no click at all. It reports poorly.

So the numbers say to move money toward capture and away from creation. Each individual move is justified. The reported efficiency improves the whole time. And a year later the volume of people searching the brand has stopped growing, the retargeting pools are shrinking, and the activity that would have filled them was cut on the strength of a measurement that could never have seen its contribution.

The defence is to start from what each piece of activity is for rather than from what it reports.

Creating demand. Reaching people who are not looking yet. Reports worst, is measured worst, and is what fills everything downstream.

Capturing demand. Search, shopping, and anything that meets an expressed intention. Reports best and is bounded by how much demand exists.

Recovering and retaining. Retargeting, lifecycle, existing customers. Efficient, small, and downstream of the first two.

Sizing those three against the business's growth needs produces a starting shape that the numbers can adjust within. An account growing fast needs more in the first bucket than its reports will ever justify; an account harvesting a mature market can defensibly weight the second.

The number that matters for an allocation decision is what the next dollar produces, not what the average dollar has produced.

Those differ, and they differ most where it matters. The cheapest conversions in any channel are bought first, so the marginal return falls as spend rises. A channel reporting an excellent average may have exhausted its cheap volume and be returning very little on its next increment, while a channel with a worse average may have room.

Averages conceal this completely. The only way to learn the marginal curves is to move budget in deliberate increments and measure what happens, which is slower than reading a report and is the difference between an account that scales and one that optimises itself into a corner.

Three things routinely justify a share that the reporting will not support.

Incrementality. A channel's reported return and its incremental return can differ by a factor of several, and the gap is widest exactly where the reported figures are best. An account that has tested knows its ratios; one that has not is allocating on claims.

Strategic commitments. Entering a market, supporting a launch, maintaining presence against a competitor. These have reasons outside the objective and a share that reflects them.

Constraints. A committed budget, a contractual minimum, a channel the business is obliged to run. These are not up for optimisation and should be set aside before the discretionary money is divided.

Set the shape by job, sized against what the business needs to grow.

Reserve the committed and strategic amounts before anything is optimised.

Divide the discretionary remainder using reported returns adjusted by whatever incrementality evidence exists.

Move in increments and measure, rather than reallocating in large steps on a single period's figures.

Review on a cadence — monthly or quarterly — rather than in response to a bad week, because weekly figures are noisy and churn costs more than it gains.

Refresh the incrementality evidence annually on the largest lines, because the ratios move as creative, audiences and the market change.

The most common allocation failure is not a wrong split. It is a split that nobody has revisited, carried forward each quarter with small adjustments, whose original reasoning nobody remembers.

It survives because it is never obviously wrong in any single period, and because the work of rederiving it is unrewarded. Writing down why each share is what it is, once, turns it into something a successor can argue with rather than inherit.

A distinction worth keeping clear: allocation between channels is a decision somebody makes, and allocation inside a platform increasingly is not.

Where a campaign pools its budget across the containers beneath it, the platform distributes continuously on predicted value. That is generally better than a person reviewing weekly, and it cannot see the reasons a person has — a committed amount, an audience being kept warm, a market with strategic weight.

So the useful division of labour is that the advertiser allocates at the level where reasons the platform cannot see apply, and the platform allocates below it. Structuring the account so those levels line up is what makes the arrangement work, and it is why structure and allocation are the same conversation.

Moving budget is not free, and the cost is not obvious in a spreadsheet.

A material budget change resets what an automated strategy is optimising against and produces a period of unstable delivery. An account reallocating monthly on small differences pays that cost repeatedly and spends a meaningful share of the year in transition.

Two practices reduce it. Move in increments rather than in swings, so changes are absorbed rather than triggering a reset. And set a threshold below which a difference is not worth acting on, which forces the question of whether the observed gap is signal or noise before anybody moves money on it.