Splitting the budget across campaigns: the marginal method, the rules and the calendar

You have a total for the month. You have eight campaigns. How much goes to each one? The usual answer ("roughly the same as last month") leaves money sitting in flat campaigns and starves the ones that are growing. This lesson gives you the method — the same one the Suite's script applies every night — to do it with data, and in steps.

The method in six steps

1. Ring-fence first. Set aside whatever is not allocated on performance:

2. Judge each remaining campaign (module 2, the judge): it meets, beats or misses its business goal, with enough volume behind it.

3. Measure its capacity to absorb: lost IS to budget. With no lost IS to budget, more money will not get spent (or will get spent badly).

4. Estimate the marginal (lesson 2): does the next tranche perform?

5. Allocate by priority:

6. Balance the total and translate to daily: monthly budget ÷ 30.4. Write down the split and the reason for each change.

Steps, not jumps

Each cycle (weekly or fortnightly) moves 10-20% at most per campaign. Three reasons: Smart Bidding keeps learning without resetting, you walk along the curve instead of jumping into the flat zone, and estimation errors get corrected before they cost money. An "optimal" split applied in one go usually performs worse than an approximate one applied in steps.

Seasonality

If the monthly total changes (sales, summer, Christmas), the split changes first: seasonal campaigns get their share one or two weeks before the peak (learning needs a run-up), and give it back once it is over. Two supporting tools: last year's history (which campaigns actually went up) and Smart Bidding's seasonality adjustments for short peaks (module 2). Never the same split all year if demand is not the same all year.

Shared budgets: when

For homogeneous campaigns (same economics, same goal, in a portfolio), a shared budget lets Google handle the daily split between them. It saves you from having five regional campaigns where the budget runs out in some and goes unspent in others. Not for mixing brand with generic, or different networks.

Mistakes that throw the split off

• Allocating by historical spend ("it has always had €500").
• Cutting the ones that beat target because "they are doing fine".
• Raising the ones that miss target "to see if they pick up".
• Moving everything at once and not knowing what caused what.
• Having no testing reserve and funding the new stuff out of what works.
• Forgetting to translate to daily: a badly divided monthly total runs the month dry by the 25th.

An example

Total €6,000/month. Ring-fenced: brand 400, remarketing 300, testing 500. That leaves 4,800 for five generic campaigns. Verdicts: A beats target with 45% lost IS to budget (→ +20%), B meets target with no lost IS (→ leave it), C misses target (→ −15%, diagnose), D beats target with no lost IS (→ leave it, it is at its ceiling), E has no data (→ leave it). About €400 moves from C to A; the rest stays. Next cycle, recalculate.

💡 Ninja trick: this method is the budget module of Smart Bidding (SBNS): configurable reserves per campaign (minimums and maximums), a nightly verdict against the business goal, lost IS and CPA trend as capacity and marginal, moves of 10-20% with a wait between cycles, and a record of every one. Plus a golden rule of engineering: if the script fails, it touches nothing — the previous split stays in force.

What you should remember

• Ring-fence (brand, remarketing, testing, commitments) before allocating on performance.
• Allocate by verdict × lost IS (budget) × marginal, in steps of 10-20%.
• Seasonality: move before the peak and give it back afterwards.
• Monthly total ÷ 30.4 = daily; write down every change and its reason.