Pega Marketing Consultant cheat sheet

NBA arbitration selects one action per customer per channel per direction (Inbound or Outbound) by evaluating eligible actions through the decision strategy and returning the highest-ranked result.

An issue tree is a hierarchy of Business Issues → Groups → Issues; each leaf Issue holds actions, and the top-level Issue (usually Sales) is what an NBA strategy references.

Applicability (eligibility) is evaluated before suitability: an action that fails its applicability condition is excluded and never scored, so eligibility rules should be cheap and deterministic.

Propensity is the likelihood a customer accepts an action, produced by a predictive model (Adaptive Model, Scorecard Model, or third-party PMML); it is one input to the final score, not the score itself.

Final score is a weighted combination of propensity, value, and levers; the Value component is typically propensity × action value (margin or expected revenue).

Levers (business levers) adjust ranking at run time by direction: a positive lever boosts an action, a negative lever suppresses it, and a lever with no matching condition has no effect.

Engagement policies (contact policies) limit how often a customer is contacted: per-action and per-group limits are evaluated against the customer's contact history within a defined period.

Suppression and fatigue rules run in the strategy as filters; the Contact Policy shape enforces volume constraints, while suppression conditions exclude specific customers or actions outright.

The NBA Designer produces a Next-Best-Action decision that can be exposed as a real-time container (e.g., web mashup or API) or run in batch for outbound audiences.

A decision strategy is a directed graph of shapes: Data, Strategy, Decision, Proposition, and Import shapes are connected by flows; execution begins at the Start shape and ends at a Result shape.

The Strategy shape (Group By) partitions the incoming customer set into groups by a property and runs its inner strategy once per group, so aggregates and rankings are computed per segment.

The Decision shape evaluates a decision table or decision tree; first-match and all-matches evaluation modes change whether one or many rows can fire.

The Proposition shape filters, ranks, and selects from a proposition data set; it is the standard way to apply eligibility, suitability, and arbitration inside a strategy.

Aggregate shapes compute counts, sums, and averages over a group (for example, number of contacts in the last 7 days) and are commonly paired with Group By for contact limits.

Strategy results are typed: a Result shape must declare a component (for example, a Page or a Proposition) that matches the downstream consumer's expectation.

Test strategies with the built-in Test capability using a sample customer page; the trace shows each shape's input and output so you can isolate where an action was filtered or re-ranked.

Strategy reuse is achieved through sub-strategies (Strategy shapes referencing another strategy) rather than copy-paste, keeping arbitration logic in a single maintained place.

Proxy strategies expose a decision to other applications; the proxy defines the input and output contract and decouples the caller from the internal strategy graph.

The Customer Decision Hub application is built on the Pega Platform; the decisioning stack includes Data Flow, strategies, predictive models, and the NBA Designer, each configured as ruleset components.

Data flows (Data Flow and Data Flow with Processing) move and transform customer and interaction data; real-time data flows run on each interaction, batch data flows run on a schedule.

A Data Set (formerly Data Set / Data Page) is the source and sink abstraction: stream, database, and repository data sets connect the decisioning layer to external systems.

Predictive models are created in Prediction Studio; Adaptive Models learn continuously from outcomes, while Scorecard Models are built from historical data and refreshed on a schedule.

Outcome capture (the Outcome shape or a response stream) records the customer's response to an action; without outcomes, adaptive models cannot learn and propensity stays flat.

The Proposition Management landing page holds propositions, treatments, and their metadata; a proposition is the business offer, a treatment is the channel-specific presentation of it.

Contact history (Interaction History) stores past contacts and outcomes; it feeds contact policies, suppression, and model learning, so its data set must be populated reliably.

Role-based access is configured through access groups and privileges (for example, Strategy Designer, Model Manager, and Proposition Manager) so each persona can only change their layer.

Revision management and rule check-in/check-out apply to decisioning rules; use branches and rulesets to promote strategies and models between development, staging, and production.

The reference architecture layers are: source systems (CRM, data warehouse) → data ingestion (data flows/data sets) → decisioning (strategies, models, NBA) → channels (web, mobile, call center, outbound) → feedback (outcomes back to Interaction History).

Real-time decisioning is invoked synchronously per interaction and must meet latency budgets; keep the strategy graph lean and precompute heavy aggregates in batch where possible.

Batch decisioning runs the same NBA strategy against an audience to produce outbound lists; it reuses the strategy but is not latency-sensitive.

The delivery lifecycle is design → build in a development environment → test with sample data → promote through staging → production, with revision management and rollback via ruleset versions.

Non-functional requirements to capture early: peak interactions per second, acceptable decision latency, data freshness (real-time vs. hourly), and availability targets for the decision service.

Security and compliance: mask PII in data sets, restrict access to Interaction History, and document consent and contact-preference handling in the strategy design.

Monitoring covers model performance (propensity drift, AUC), strategy health (fallback rate, filtered-out rate), and system health (data flow failures, latency), each with an owner and alert threshold.

Integration contracts (proxy strategies or APIs) define the request and response schema; version them so channel teams can upgrade independently of decisioning changes.

A solution delivery plan sequences proposition onboarding, model training, channel integration, and outcome feedback so each stage produces data the next stage needs.