Reallocating a marketing budget with explicit rules.
Evidence status
What was reproduced, and what it does not prove.
The fixture and the Python run verify that minimum and maximum shares are respected, the robust marginal calculation and the rescaling of the educational allocator's constraints. They prove neither the persistence of a signal nor the gain of a market reallocation.
- Technical specification of the budget allocator, specification in French (opens in a new tab)
v1, 31 August 2026 · CC BY 4.0 · specification in French · Method document written by marketing-mix.net. The figures are illustrative: they are not market measurements.
Key distinctions
Three conditions before calculating.
A signal is not yet a decision
A fall can come from noise, from a conversion delay, from a stockout or from a change of measurement. The rule sets a reference, a window and a threshold before any move.
The mobile tranche is bounded
Commitments, minimum presence, learning in progress and capacity determine the share genuinely movable. The total budget of a channel is never assumed fungible.
Reversal is prepared
The review date, the stopping signal and the restoration reserve are defined before execution. A rule with no documented exit is not reversible.
Method
Writing a reallocation rule in four decisions
- 01Qualify the trigger
Choose a measure aligned with the decision — marginal contribution, incremental cost, saturation, capacity or availability; set the minimum period, the reference and the uncertainty.
- 02Compute the prudent tranche
Compare the contribution lost at the source with the one expected at the destination; retain a prudent bound, then deduct the transition costs and the continuity minimums.
- 03Test the constraints
Check coverage, learning volume, stock, service, calendar and correlations. A blocking constraint forbids the move even if the central gain is positive.
- 04Log the revision
Record amount, assumptions, owner, effective date, observation window and the conditions to continue, stop or restore.
The register of reallocation rules
Every move connects an observed signal to a window, a tranche, a guardrail and a review decision.
| Rule | Signal | Window | Threshold | Tranche | Guardrail | Review | Decision |
|---|---|---|---|---|---|---|---|
| R01 | Marginal return | 4 weeks | +0.30 | €40k | Stable reach | D+14 | Shift |
| R02 | Net prudent gain | 6 weeks | > €0 | €40k | Observed range | D+42 | Extend |
| R03 | Availability | 7 days | < 95% | €0 | Restore if out of stock | D+7 | Block |
| R04 | Net prudent gain | 6 weeks | < €0 | −€40k | Reserve available | D+42 | Restore |
Moving €80k in two steps rather than €200k at once
Illustrative example: channel A has a prudent marginal return of 0.10 and channel B of 0.62. A has €500k but must keep a floor of €420k. The transition cost is estimated at €6k.
The maximum tranche is 500,000 − 420,000 = €80,000. The gross prudent gain is 80,000 × (0.62 − 0.10) = €41,600. After €6,000 of transition, the prudent expected gain is €35,600.
The rule authorises €40k immediately and then €40k after review. It blocks any further transfer as long as B's response and A's coverage have not been re-estimated.
Acceptance conditions
What must be true to act.
- 01
The trigger uses an incremental or marginal measure adapted to the horizon.
- 02
The amount respects minimums, maximums, commitments and observed ranges.
- 03
The gain stays positive under a prudent assumption after transition costs.
- 04
A review date, a stopping threshold and a condition for reverting are written before execution.
Limits
What this analysis does not prove.
- A rule does not correct a biased measurement nor a poorly identified response curve.
- Interactions between channels can invalidate a calculation based on separate effects.
- Brand effects and long delays can require dedicated floors.
- Changes that are too frequent disturb learning and prevent evaluation.
References