AI optimisation of the digital marketing campaign mix
Digital marketing budgets are rarely ‘won’ by a single channel, but rather by the combination of channels: Search, social, display, retail media, affiliate, video, CRM and marketplaces all influence one another. At the same time, there are strict constraints imposed by budget, brand safety requirements and concentration risks. The key question is: How should the budget be allocated to maximise incremental revenue – without ignoring cannibalisation effects?
Objective
Maximising incremental revenue whilst adhering to an overall budget and defined brand safety limits. It is not ‘spend’ that is optimised, but the additional revenue causally generated by the campaign mix.
Assessment Inputs
- Expected conversion lift per channel (incremental, not just last-click)
- Cost per impression / action (CPM, CPC, CPA, CPO)
- Brand safety score per channel (0–1)
- Cannibalisation risks between channels (overlaps, diminishing returns, substitution)
Evaluation and selection mechanism
The budget shares per channel are modelled as decision variables (budget allocation across all channels). A non-linear objective function is optimised, which takes into account both revenue contribution and cannibalisation:
max Σ rᵢ xᵢ − λ · Cannibalization(x)
Here, rᵢ represents the expected incremental revenue contribution per budget share in channel i. The term Cannibalization(x) captures overlaps and substitution effects (e.g. social vs. search, display vs. retail media), and λ controls the extent to which these effects are penalised in the optimisation.
Constraints
- Total Spend : ≤ 1,000,000 USD
- Concentration limit : no single channel > 40% of the total budget
- Brand Safety : weighted average ≥ 0.85
Result
- Optimal budget allocation across channels with maximum incremental revenue
- Explicit compliance with spend, channel and brand safety limits
- Transparent trade-offs: additional revenue vs. cannibalisation risk
- Reproducible decision-making logic for CMOs, performance teams and finance
Technology
StratePlan calculates the optimal budget allocation subject to constraints, including non-linear objective functions and interaction effects between channels. Marketing Mix Modelling (MMM) provides the impact functions and response curves that quantitatively model incremental effects, diminishing returns and cross-channel interactions and incorporate them into the optimisation.
Common Patterns Across Cases
Assessment
Qualitative and quantitative factors are converted into comparable scores – using scales, evaluation models or structured expert assessment. The aim is to establish a consistent, decision-ready evaluation basis.
Ranking
Elements are prioritised. However, ranking is rarely the final decision. In complex environments, prioritisation is frequently embedded directly into a combinatorial optimisation process in order to systematically account for interactions and constraints.
Group Selection
The final selection goes beyond a simple ‘Top-k’ approach. StratePlan solves structured selection problems such as knapsack, portfolio or scheduling models and calculates the optimal combination under real-world constraints.
Constraints
Constraints reflect real-world scarcity: capital, time, resources, risk appetite, regulatory requirements, strategic mandates or sustainability requirements. They are an integral part of the decision-making logic.
Technologies
Hybrid use of MCDA methods (e.g. AHP, TOPSIS) for structured evaluation combined with StratePlan for constraint-aware group or portfolio selection.
These cases demonstrate how StratePlan evolves decision-making processes from pure ranking to intelligent, constraint-aware portfolio construction. Evaluation data is translated into actionable, optimised group decisions – aligned with financial, strategic and sustainability-related objectives.
The underlying core logic – structured evaluation → quantitative prioritisation → constrained group selection – scales across different sectors and is adapted in each case to domain-specific success metrics and constraints.
