Decisioning Data and Strategy Components
This domain tests the data foundation that feeds decisioning and the strategy shapes you assemble to produce candidates, propensity and value.
Data foundation
- Customer profile – typically Data-Party classes holding demographics, holdings and calculated properties.
- Interaction history (IH) – impressions, clicks and accepts recorded per customer; this is the feedback that models learn from.
- Data flows / data sets – move and transform data into the Decision Data Store (DDS); streaming can update context near real-time.
- Context Dictionary – defines the customer context (often a data page) that strategies read at decision time.
- Subject ID / primary fields – identify the customer consistently across data sets and IH.
Strategy shapes worth naming
Decision Table, Decision Tree, Scorecard, When, Filter, Group By, Sort, Prioritize, Map Value, Set Property, Sub-strategy, Adaptive Model, and proposition/action data shapes. Filters reduce rows early; Group By limits how many actions leave a branch; Prioritize and Sort order the output.
Worked example
A strategy for inbound web must exclude customers with fewer than 30 days' tenure, assign a value by segment, drop ineligible actions, keep the best two per issue and score the rest with an adaptive model. Order matters: filter tenure first (cheapest reduction), map value by segment, filter ineligible actions, Group By issue to cap output, then call the adaptive model only on survivors, and finish with Prioritize. Reasoning: scoring before filtering wastes model calls and lets ineligible actions distort the output.
Traps that cost marks
- Placing the adaptive model before filters, so ineligible actions get scored.
- Forgetting that strategies must output the action plus propensity and value for arbitration.
- Mismatching subject IDs between data sets and interaction history, so outcomes never attach.
- Assuming Group By sorts; it caps, it does not rank.