Action arbitration

This video explains the concept of action arbitration in Pega Customer Decision Hub.

Pega Customer Decision Hub combines analytics, business rules, customer data, and data collected during each customer interaction to create a set of actionable insights that it uses to make intelligent decisions. These decisions are known as Next-Best-Actions. Every Next-Best-Action weighs customer needs against business objectives to optimize decisions based on priorities set by the business manager.

U+, a retail bank, wants to avoid offering random actions to its customers, so it uses Pega CDH to rank and select the next best actions for every customer. All the actions from the Bank's offer move through several steps that whittle them down until the Next Best Action is selected.

The process starts with the catalog full of actions that the organization can offer to its customers. These actions belong to various issues and groups.

The first step is to apply the Engagement Policies: Eligibility, Applicability, and Suitability conditions.
Eligibility rules are used to determine whether customers are eligible for an action. Users must meet the condition for CDH to consider the action eligible. An example of an eligibility condition is the customer's age. A customer can be offered a credit card only if they are 18 years old or older.
Applicability rules are used to limit what to offer, based on a customer's current situation. For example, the bank wants to show a retention offer, instead of a credit card offer, to customers who are likely to churn in the near future.
Suitability rules are used to define whether an offer is appropriate for a customer. For example, if a customer is in debt and has a lot to pay off, the bank does not want to offer them a 30-year mortgage as it would not be empathetic.

The initial catalog of actions has been narrowed down by engagement policies.
These engagement policies are a combination of business rules, for example, certain age requirements for a loan or some product compatibility rules, and predictive models which help predict customer behavior as customer churn or probability of default. These predictive models are created by data scientists in Pega, or other tools familiar to data scientists like Python, R, and H2O.AI.

Before being presented to customers, the remaining actions go through a final stage known as the arbitration stage. This stage focuses on selecting and prioritizing the best actions that are most relevant for the customer at the present time. Arbitration balances customer relevance with business priorities using four components: Propensity, Context Weighting, Business Value, and Business Levers, each represented by numerical values.

Propensity is the predicted likelihood of positive behavior, such as the likelihood of a customer accepting an offer.
The propensity value is calculated by AI and it is the foundation of the arbitration process.
The higher the likelihood of a customer accepting an offer, the higher the Propensity value for that offer. For Example, when a customer is eligible to receive a Standard Credit Card and clicks the Standard Credit card offer on the Bank's website the propensity of this action increases.

The AI is driven by adaptive models, self-learning mathematical models that use machine learning to make calculations. Adaptive models are a sub-group of predictive models that continuously adapt to new data as it becomes available. This means that the models are updated incrementally in real-time based on incoming data.
The Customer Decision Hub is configured to calculate the propensity for each treatment. For the web treatment, Web_Click_Through_Rate is the out-of-the-box adaptive model. It predicts the likelihood of a customer clicking on a banner, which is considered a positive behavior.

Context Weighting allows Pega Customer Decision Hub to consider the situational context for each action. For example, if a customer contacts the bank to close their account, the highest-priority action is to ensure that the customer is retained, even if they are eligible for other offers.

Business Value enables you to assign a financial value to an action and prioritize high-value actions over low-value ones. This value is typically normalized across Issues and Groups. For example, a 25-year mortgage is more profitable than a 15-year mortgage. So, in a situation where a customer is eligible for both plans, the 25-year mortgage will be ranked higher because of its higher business value.

Business Levers allow the business to assert some level of control over the prioritization of actions defined within the system. Levers are used to manually nudge Customer Decision Hub toward Next-Best-Actions based on external factors. For example, the recommended Next-Best-Action might be to offer a credit card to a customer when they visit the home page. However, to meet a business goal, the Mortgage Line of Business favors a mortgage offer, even if that offer is ranked a little lower on the list of possible actions.

Pega AI considers all four components to select the top offer for a customer. A simple formula P*C*V*L is used to arrive at a prioritization value, which is used to select the top actions. The action with the highest priority will be presented to a customer.

In the current example, there are 5 actions that made it until the arbitration phase.

Assume the AI generates a propensity value for each of the actions which is a value between 0 and 1.

Each action has a context weight set, which is a percentage value between 0% and 100%, business value in dollars, and business lever value in a percentage between 0% and 100%.

Using the priority formula PCVL, with all of these values taken into account, we end up with a computed priority. Finally, the action with the highest priority is selected. In this case, it is a Standard Credit Card.

In summary, arbitration is used to select and prioritize the best actions that are most relevant for the customer at the present time. Arbitration balances customer relevance with business priorities using the formula P*C*V*L. The top actions are selected based on the result of this multiplication – the prioritization value.