The Decisioning Design Problem
Decisioning logic is the operational core of personalisation — the set of rules or model outputs that determine which content component is shown to which audience member in which context. Without explicit decisioning design, personalisation systems default to one of two failure modes: over-personalisation (showing radically different content to different audiences without understanding why the difference is appropriate) or under-personalisation (showing broadly similar content to all audiences with superficial variation at the surface level).
Rules-Based vs. Model-Based Decisioning
Rules-based decisioning: Human-authored if-then logic that maps audience attributes to content selections. "If audience segment = enterprise financial services AND journey stage = evaluation AND topic interest = risk management, then show content component X." Rules-based decisioning is interpretable, controllable, and fast to implement. It degrades at scale — maintaining hundreds of rules is operationally intensive, and rules cannot capture the nuance of individual behaviour patterns.
Model-based decisioning: Machine learning models that learn from historical content performance data to predict which content will be most relevant for a given audience profile. Model-based decisioning scales well and captures pattern complexity that rules cannot represent. It is less interpretable, requires more data to train, and can produce recommendations that are difficult to explain or override.
Hybrid decisioning: The most practical architecture for most organisations. Rules govern the constraints — content that must be shown (legal disclosures, product requirements), content that must not be shown (competitor references, deprecated claims), and audience segments where model recommendations are not yet trusted. Models govern the optimisation — within the rule constraints, selecting the content most likely to produce the desired outcome.
Key Takeaways
1. Decisioning logic must be explicitly designed — personalisation systems without designed decisioning logic default to under-personalisation or over-personalisation.
2. Hybrid decisioning — rules for constraints, models for optimisation — is the most practical architecture for most organisations at most stages of personalisation maturity.
3. Decisioning logic must be testable — A/B testing, holdout groups, and performance attribution are the governance mechanisms that validate decisioning quality and enable continuous improvement.