← Archimedes03 / 04
Proprietary engine — module 03

Budget allocation

Given a budget, Archimedes returns the split that maximises projected return — and shows what the second-best split costs you.
Budget allocation — Archimedes module at Atlas Marketing
Allocation runs on the response curves produced by the forecasting module. Because each channel has a saturation point, the optimal split is rarely the one that feels intuitive: it usually means capping a channel that looks efficient at low spend and funding one that is under-invested rather than under-performing.
The optimiser is constrained, not theoretical. Minimum brand-building share, contractual commitments, seasonal blackouts, retail media requirements and creative production capacity all enter as constraints, so the output is a plan that can actually be executed in Athens next Monday.
We rerun it monthly against actuals. If the market moves, or a channel's cost per acquisition drifts, the plan is reforecast rather than defended.

Specification

Cadence
Quarterly plan, monthly reallocation
Minimum budget
Meaningful from ~€10k / month media
Calibration
Geo holdout or lift test per major channel
Turnaround
3 working days per scenario set

Data inputs

  • Total budget and any fixed commitments
  • Historic spend and results by channel
  • Channel response curves from the forecast module
  • Margin by product line
  • Production and creative capacity

How it runs

4 stages
  • 01

    Curve fit

    Per-channel response curves estimated from spend history and validated against geo or holdout tests where available.

  • 02

    Constrain

    Contracts, minimum brand share, seasonality windows and production capacity encoded as hard limits.

  • 03

    Optimise

    Constrained optimisation across channels and weeks to maximise projected contribution at the given budget.

  • 04

    Compare

    Optimal plan shown against last year's plan and the client's proposed plan, in projected revenue terms.

Methods inside the model

4 methods
  • 01

    Constrained optimisation

    Maximises projected contribution subject to budget, contract and brand-share constraints — not a spreadsheet of percentages.

  • 02

    Marginal-return balancing

    Spend is moved until the last euro in each channel returns the same, the classical condition for an efficient split.

  • 03

    Incrementality testing

    Geo holdouts and conversion-lift studies used to calibrate curves against causal, not correlated, results.

  • 04

    Scenario deltas

    Every alternative plan is priced: 'this split costs €X in projected revenue' rather than 'this split is worse'.

What it produces

  • /Quarterly budget split by channel and week
  • /Marginal return table with saturation points
  • /Projected revenue delta versus current plan
  • /Reallocation recommendations, monthly

Honest limits

  • Without at least one incrementality test, curves rest on correlation and the split is directional rather than exact.
  • Brand-building spend is protected by constraint, not by the optimiser — short-horizon maths will always underweight it.
Next moduleSignal monitoring →
Talk to us about Archimedes →