Pega CPDC flashcards

33 cards. Read the term and the definition here. The deck at the top of this page is optional.

Next-Best-Action

A decisioning approach that evaluates all eligible actions for a customer at an interaction, scores them, and selects the action with the highest expected business value for delivery.

Proposition

The offer content presented to a customer, such as a product recommendation or message. A proposition is associated with an action and rendered through a channel-specific treatment.

Treatment

The channel-specific presentation of an action, defining how the offer appears (for example, web banner, email, or call-center script). One action can have multiple treatments.

Real-time container

The runtime component that captures the customer context at an interaction and invokes the decisioning strategy to return the next best action within the response-time budget.

Interaction history

The record of decisions made and customer responses (impression, accept, reject). It feeds adaptive model learning, arbitration analysis, and performance reporting.

Strategy shape

A building block in a strategy, such as Data Import, Filter, Group By, or Set Property. Shapes pass page lists and the final shape produces the strategy result.

Data flow

A Pega component that moves data between the operational datastore and external sources. Real-time flows serve individual decisions; batch flows run on a schedule.

Operational datastore

The PegaDATA schema that stores customer, interaction, and decisioning data used at runtime. It is separate from the rules database that stores application rules.

Predictor mapping

The configuration that links a customer property to an adaptive model predictor. The property must be populated at decision time or the model loses signal.

Customer profile class

The class holding customer attributes used in decisioning, typically Data-Customer or an application subclass. Strategies import from it to evaluate eligibility and score actions.

Strategy result

The output page list of a strategy (pxResults) containing the scored, filtered candidate actions that downstream arbitration and delivery components consume.

Engagement policy

The NBA Designer tab that defines which actions are applicable to a customer, using audience conditions and the issue/group hierarchy to filter the catalog.

Issue and group

The two-level hierarchy that organizes the action catalog. Issues are business areas and groups are subcategories; the hierarchy drives eligibility and reporting.

NBA Designer

A guided authoring interface for business users to configure engagement policy, arbitration, and contact policy, producing a decisioning strategy without rule-level coding.

Contact policy

Rules that limit how often and through which channels a customer can be contacted, including suppression, fatigue limits, and channel caps applied during decisioning.

Action weight

A business value assigned to an action that reflects its importance or revenue. It multiplies with propensity in arbitration to produce the action's expected value.

Arbitration

The process that ranks eligible, scored actions and selects the winner(s) according to priority, propensity, and business value, subject to contact policy.

Arbitration formula

The configurable expression that combines propensity, value, and priority to rank candidates. The default multiplies propensity by value and uses priority as a filter or tiebreaker.

Priority

An integer assigned to an action that can force it ahead of higher-propensity actions. Priority is often applied before value comparison as a filter or ordering rule.

Top-N selection

The arbitration setting that returns the highest-ranked N actions instead of a single winner, enabling bundles or fallback offers when the primary action is unavailable.

Eligibility filter

The arbitration stage that removes actions the customer cannot receive, based on audience conditions, suppression rules, or missing data, before scoring and ranking.

Null propensity

A missing model score for an action. It typically causes the action to be dropped or scored as zero, so predictor mapping must be verified before go-live.

Adaptive model

A self-learning model that updates predictor weights after each outcome. It starts learning immediately and predicts the probability of a positive customer response.

AUC

Area under the ROC curve, the standard adaptive model performance metric. Values near 0.5 indicate no discrimination; values near 1.0 indicate near-perfect ranking.

Learning rate

A setting controlling how much each new outcome changes adaptive model weights. High values adapt quickly but are noisier; low values are stable but slow to react.

Predictor binning

The process of grouping a predictor's values into bins so the adaptive model learns a weight per bin rather than per raw value, improving generalization.

Minimum responses

The number of outcomes needed before an adaptive model's AUC is meaningful, commonly around 100-200 positive responses. Below that, scores are unreliable.

Adaptive Model report

A report showing an adaptive model's AUC, response counts, and predictor performance, used to monitor model health and decide when to retrain or adjust predictors.

Simulation testing

Running a candidate strategy or NBA configuration against a sample of historical or synthetic data to estimate outcomes and lift without affecting real customers.

Sample population

The set of customers or interactions used in a simulation. It must be representative of the target audience, or the estimated lift will be misleading.

Lift

The estimated improvement of a candidate strategy over the current production strategy, usually measured as the difference in predicted acceptance rate in simulation.

Test run

A real execution of a strategy on a limited set of live interactions, distinct from simulation, which uses historical or synthetic data and has no customer impact.

Simulation sandbox

An isolated environment where simulations execute. It does not modify production data, and its results are used for analysis rather than actual customer delivery.