Compare a historical allocation with a robust scenario under operating constraints.
Reference case
Reallocate €1.2m without increasing the budget
Decision questionHow should €1.2m be reallocated across channels under uncertainty?
Three years of data, five channels and one actionable question: where should the next euro go to raise contribution without hiding the uncertainty?
Decision taken
Commit to the reallocation in two steps.
First reduce the concentration on paid search and paid social. Progressively strengthen video, television and retail media. Moving to the target stays conditional on six weeks of validation.
+€121,400expected contribution / quarter
+6.2%central scenario
0 €additional budget
88%positive gain in the simulation
01
The problem
The budget is stable. Its marginal productivity is not.
Noroa puts 65% of its budget into paid search and paid social. Those channels still produce, but their next budget unit returns less than one placed on under-invested channels.
Optimise incremental contribution, not attributed revenue or platform ROAS.
No additional budget. An explicit minimum and maximum for every channel.
A six-week pilot step, two weeks of waiting and a decision in week nine.
02
The data
Fifteen variables tied to one precise decision.
The weekly file brings together economic outcome, media investment, price, promotions, availability, store coverage, seasonality and calibration events.
CSV · 156 ROWSWeekly data
The full series, a dictionary matching the model and the estimation-validation split.
Download the CSV ↓JSON · DECISION MODELParameters and resultsCurves, constraints, allocations, diagnostics, scenarios and decision rules.
Download the JSON ↓MARKDOWN · DECISION BRIEFDecision briefAssumptions, formulas, the allocation table, the protocol and the limits in a reusable format.
Download the brief ↓OutcomeRevenue and contribution after media
MediaSearch, social, video, TV, retail media
ControlsPrice, promotion, availability, stores, season
Validation12 weeks held back + 2 calibrated events
03
The decision model
Compare marginal returns, not historical averages.
Each channel gets a saturation curve and a confidence level. The allocation moves in €10,000 steps within the published operating bounds.
Decision model
Ri(s) = Ai × sαi ÷ (Hiαi + sαi)A is the potential, H the half-saturation point and α the shape of the curve. The observed range and the uncertainty are kept in the model file.
144 weeks of estimation and 12 weeks held back for out-of-time validation.
Mean absolute error on revenue for the weeks not used to estimate the model.
Eleven observations out of twelve fall inside the predictive interval stated at 80%.
VIDEO GEO TESTSimulated lift 5.6% · model 5.1%
RETAIL-MEDIA GEO TESTSimulated lift 8.4% · model 7.9%
These events illustrate calibration; they are not observed market performance.
04
The outcome
€270,000 moved to three under-invested channels.
The target scenario cuts no channel. It reduces spend already close to saturation and strengthens the areas where the marginal return remains higher.
ChannelCurrentConditional targetChangeMarginal return
Paid searchConfidence 88%
€440,000
€290,000
-€150,0000.7 € / €Paid socialConfidence 62%
€340,000
€220,000
-€120,0000.8 € / €Online videoConfidence 70%
€140,000
€230,000
+€90,0001.5 € / €TelevisionConfidence 66%
€200,000
€270,000
+€70,0001.3 € / €Retail mediaConfidence 80%
€80,000
€190,000
+€110,0002 € / €Total€1,200,000€1,200,0000 €1.2 M€
Weaker response from video and television, higher exposure costs.
Central parameters, constraints respected and a strictly constant budget.
Calibration confirmed and commercial availability stable across the whole period.
05
Implementation
A recommendation becomes a decision only with a stop rule.
The target is a conditional scenario, not an automatic order. It must survive a compared pilot, the reconciliation of the data and economic guardrails.
Set comparable regions, exclusions, contribution after media and stop rules before moving budget.
Use intermediate shares in treated regions without changing total budget or promotion rules.
Allow delayed effects, then check availability, price, promotions and data joins.
Move to target, hold the pilot or revert according to the published thresholds.
And availability above 95%, comparable price and promotions, no critical margin deterioration.
Hold the pilot if the prudent loss stays below €20,000 and measurement quality can be improved.
Or availability below 95%, a break in comparability, or a delivery constraint exceeded.
06
Responsible reading
This case demonstrates a method. It does not demonstrate a market.
The results are consistent with the published data and parameters. They must never be transplanted into a real company without re-estimation.
- Simulated data
No gain shown here has been observed at a real advertiser.
- Limited interactions
The curves do not estimate creative or sequential synergies between channels.
- Simplified costs
Agency, production and transformation costs are not itemised.
- Local decision
The minimums, maximums and confidence levels are specific to the Noroa case.
Your turn
Test the logic against your own constraints.
Use the allocation laboratory to compare your own channels, then take the protocol from the case to turn the scenario into a measurable decision.