Stop defending an amount; have the consequences traded off
Presenting only an amount and a central forecast shifts the debate to the trust placed in the forecaster. A decision dossier reverses the posture: it shows what several envelopes make possible, what they sacrifice and under which constraints they remain feasible. The studies reviewed do not give the right budget; they specify what the trade-off must make verifiable.[1][2][3]
The first page of the dossier then sets three scenarios against each other: baseline, proposed reallocation and tightened constraint. It answers the chief executive’s and the finance director’s questions directly: why this envelope, what is lost with 20% less, which cost can be removed, which part is committed and when must the matter return to the committee? This structure is an original decision grid, built for the use presented here and not validated as a method by the corpus.[1][3]
Allocation must integrate business constraints
The framework presented by Alibaba brings several categories of business constraints into a single optimisation paradigm. It combines response modelling with continuous or discrete allocation. Its interest for a committee lies in that coherence: constraints are not added as a footnote after the calculation, they are part of the problem to solve. The text does not, however, prove universal external replication.[1]
The authors state that this framework was applied to several scenarios at Alibaba Group. This industrial case documents the real use of a constrained approach, but it remains reported by the organisation that designed it. To reuse it, a company would have to make explicit its own units, horizons, response functions and execution capacities, rather than copying the parameters of the published case.[1]
Large scale does not remove the need for control
The Ant Group study evaluates an offline constrained reinforcement-learning method on a campaign described as reaching tens of millions of users, with more than a billion yuan of coupons distributed and redeemed. The authors then report serving the entire traffic of that campaign. These orders of magnitude establish the scale of the case, not the automatic transferability of its results.[2]
For a committee, this case justifies distinguishing historical validation, controlled testing and rollout. A method can be technically deployed at large scale without the same gains being guaranteed in another organisation. The budget dossier must therefore isolate what is observed in its own data, what is estimated and what comes from a documented external case.[2]
Final reach is not enough to evaluate a campaign
The study on viral campaigns retains several characteristics linked to reach and cost. Its scenarios, built from synthetic networks, are tested against a real network. The useful result is not a ready-made scale: it is the demonstration that an evaluation can be multi-criteria and that final reach alone does not sum up the quality of a plan.[3]
The dossier presented to the committee should consequently show a baseline, the constraints, several scenarios and a reforecast rule. It can compare reach, cost and other criteria visible in the decision model, but it must avoid reducing uncertainty to a single point gain. The budget thus becomes a governed assumption, open to revision at a stated date and threshold.[1][2][3]
Going further: funding the conditions of execution
The survey of SMEs studies innovation, governance, capital structure and human capital as intermediate mechanisms between strategy and performance. The review devoted to influencers identifies, for its part, marketing objectives, budget, product and target market among the internal selection factors. These studies do not propose a ready-made calendar. They do show, however, that an isolated amount describes neither the capabilities to mobilise nor the context of the choice.[4][5]
The plan can turn this observation into preparation work packages. Each budget scenario states the capabilities required, their owner, their lead time and the expected evidence, then connects the allocation to the objective, the product and the target market. An option that costs less in media may demand more coordination or skills; a more visible option may be poorly aligned with the target. The committee thus trades off the full cost of execution, not only the purchasing line.[4][5]
Three levels of evidence to defend a budget
The publications gathered here do not play the same role. The Alibaba framework documents an allocation architecture and a controlled test; Ant Group describes a very large-scale system with operating results; the viral-campaign study compares scenarios against several criteria. A budget dossier gains from keeping these levels separate: method, observed result and governance rule. Merging them into a single return percentage would hide the contexts, units and constraints that give the reported figures their meaning.[1][2][3]
The committee can therefore require three columns. The first describes what the source establishes in its own context. The second states what the company observes in its data. The third sets out the assumption that justifies the amount requested. This presentation avoids using an external performance as an internal forecast. It also makes a reforecast possible: if the observed indicator departs from the stated threshold, the budget is reallocated without rewriting the initial assumption retroactively. The discipline bears as much on traceability as on optimisation.[1][2][3]
The Alibaba case documents a gain, not a promise
In the Taopiaopiao experiment described by Alibaba, the proposed approach improves sales by more than 6% while spending 40% less on marketing. This result is especially useful because it links a commercial effect and a spending constraint in a real test. It must not, however, be turned into a general benchmark: the platform, the response function, the allocation rules and the test period are specific to the case. The figure supports the value of experimenting with a reallocation, not the level of gain another company will obtain.[1]
To be defensible, an adaptation should announce before the test the population, the duration, the reference budget, the allocation rule, the sales indicator and the stopping threshold. It should also keep a control variant and document execution costs. The committee would then not be voting a return borrowed from Alibaba; it would be voting a learning budget with a decision rule. If the local gain does not appear or if the constraints change, the allocation is revised. The external case serves as a documented precedent, not as a forecast ready to copy.[1]
The Ant Group case links activity and coupon cost
Ant Group reports a 0.6% rise in weekly activity over seven days and a 0.8% fall in coupon cost. The interest of this result lies in the joint reading of a usage metric and a cost metric. It complements the scale of the case, which concerns tens of millions of users and more than a billion yuan of coupons distributed and redeemed. These data document the deployment as reported by the authors. On their own, they do not establish the incremental profitability of a campaign in another commercial environment.[2]
This twin indicator shows why a committee must refuse an isolated metric. A cost reduction can degrade activity; a rise in activity can cost more than its value. The dossier must therefore make explicit the decision function, the horizon and the safeguards. It must also distinguish statistical result, economic value and deployment capacity. The study reports that the authors observed both stated variations in their system. The conclusion on value creation remains to be established with the units, margins and behaviours specific to the company deciding.[2]
Maximising reach can degrade the trade-off
The multi-criteria study shows that a strategy reaching 100% coverage is not always retained, notably when it requires more initial seeds or incentives. This result gives concrete form to the budget trade-off: the maximum of one indicator can be dominated by a solution more balanced across several dimensions. The scenarios come from synthetic networks and are then verified in a real network. The work does not set the committee’s weights; it demonstrates the usefulness of making reach, cost and mobilised resources visible at the same time.[3]
A budget grid can translate this principle without imitating the network model. For each scenario, it shows expected reach, total cost, the scarce resource, lead time, risk and reversibility. Weights are announced before the trade-off and a sensitivity analysis shows whether the ranking changes with them. If the maximum-reach solution consumes a disproportionate share of budget or operational capacity, the trade-off becomes explicit. This method is the article’s original contribution; it remains to be calibrated on internal data.[3]
What these cases do not allow anyone to promise
The Performance Max rollout described by Deezer and Artefact provides an operational example distinct from the scientific experiments. The teams started with tests in the United States and Latin America, continued in the United Kingdom and Germany, then extended the set-up. Google and the agency report a 28% rise in subscriptions on the web journey and a cost per subscription 15% lower than on the app journey. These figures are self-reported and unaudited; the contribution of the case is therefore the sequence of preparation and generalisation, not a promise of results.[6][7]
The two company cases are described by their authors and the third concerns viral campaigns in a network. None provides a universal return, a comparable currency or a rule applicable to any portfolio. They make it possible to defend the structure of budget reasoning, not to announce a company’s future result. Their scope remains limited to the contexts, populations and decisions actually studied.[1][2][3]
Original decision grid
This grid turns the analysis into questions for action. It is not presented as a scientific result.
| Objective | Write the expected economic decision, not only the channel being funded. |
|---|---|
| Baseline | Present the current budget and the scenario without reallocation. |
| Assumption | Name the critical assumption that links spend to contribution. |
| Options | Compare at least a baseline, a proposal and a constraint. |
| Full cost | Add up purchasing, production, agency, technology, data, internal time and dependencies. |
| Contribution | Show the expected indicator with its unit and its range. |
| Uncertainty | Show a distribution or several scenarios, never a single point gain. |
| Capacity | Make FTEs, scarce skills and coordination workload visible. |
| Dependencies | Name the IT, data, sales and supplier prerequisites. |
| Risk | Distinguish outcome, execution and irreversibility risk. |
| Reversibility | State what can be stopped, deferred or reallocated. |
| Evidence | Separate external fact, internal observation and decision assumption. |
| Threshold | Define the green zone, the watch zone and the escalation trigger. |
| Decision | State exactly the option submitted to the committee. |
| Review | Set the date and the rule that trigger the reforecast. |