Decision strategy design and development
Domain focus
This domain tests hands-on Strategy Canvas development: assembling components into a runnable decision strategy, binding it to proposition data, and proving the result with the strategy test tools.
Core rules
- A strategy starts with Proposition Data, which supplies candidate actions from the decision data flow. Without it, downstream shapes have nothing to rank.
- Sequence matters: filter ineligible candidates early with a When rule or Filter shape; group by issue, group, or action; compute propensity with an Adaptive Model or Scorecard; set value or priority properties; then prioritize.
- Use the context dictionary to reference customer, interaction, and proposition properties. Map model inputs explicitly; a model with missing predictors returns unreliable propensity.
- Keep business rules in decision tables or trees where non-technical users must maintain them, and keep strategy logic readable.
- Test with the strategy test panel and inspect pxStrategyResult; use simulation to compare outcomes across a population.
Worked example
A web strategy needs top three card offers. Proposition Data loads five actions. A Filter removes actions where the customer is not eligible. An Adaptive Model scores response propensity. A Set Property computes expected value as propensity times offer value. Prioritize ranks by expected value and keeps three. Reasoning: filtering before scoring prevents wasted model calls, and prioritization after value calculation reflects business outcome, not raw propensity.
Traps
Omitting Proposition Data; grouping on a property that is not on the proposition; filtering after Prioritize and wondering why counts differ; and testing only one customer instead of checking a sample.