Enhance decision strategies with predictive models built on customer interaction data and let Pega Customer Decision Hub™ bring even more relevance to every customer engagement. Build models using Pega's machine learning capabilities, import models built with third-party tools, and incorporate the latest AI algorithms into the Pega AI engine by connecting to the Google AI Platform and Amazon SageMaker machine learning services.
This video will describe the use of predictive models to enhance the next best actions that Customer Decision Hub generates.
Next best actions balance customer relevance and business priorities by selecting the actions with the highest priority.
The priority is calculated by multiplying the values for propensity, context weighting, action value, and business levers.
Propensity is the likelihood of a customer responding positively to an action by, for example, clicking on a web banner or accepting an offer.
This is calculated by predictive models. In Pega, self-learning Naive Bayes models, which are generated for each action, are a key feature.
These adaptive models are automatically updated after new responses have been received and can start without any historical information because they learn on the fly.
When the use case requires a more advanced modeling technique, for example to predict customer churn or to estimate credit risk …
… Prediction Studio offers several methods to create the artifacts that represent an actual predictive model or that reference a predictive model.
The first method is to use Pega machine learning. You can import a file containing the historical customer interaction data set and build a model in Prediction Studio.
This model can then be used in decision strategies. When the decision strategies execute, the models are executed inside the Pega platform.
The second option is to import an existing model. You can build a model using a third-party tool like R or Python and export it as a PMML file.
PMML is an XML-based standard that is designed to facilitate the exchange of models between applications.
Import the PMML file into Prediction Studio and map its predictors to the fields in the customer data model.
Similarly, you can import model files that have been generated in H2O.ai. H2O is a modeling platform, and the procedure for using the generated model file is identical to that for a PMML file.
Just like with Pega machine learning models, the imported model can then be used in decision strategies.
When decision strategies using the imported models execute, the models are executed inside the Pega platform.
The third option is to reference a model on an external platform like the Google AI Platform.
Just like with Pega machine learning models, the referenced model can then be used in decision strategies.
In this case, when the decision strategy requires a prediction, a request is sent to the external model, which calculates the outcome and sends it back to Pega.
Like with the Google AI Platform, you can connect to AWS SageMaker and run your model remotely.
To summarize, you have three options for leveraging predictive models built on customer data.
You can build models using Pega machine learning, you can import models built with third-party tools, and you can use machine learning services to reference predictive models.
When the decision strategies using predictive models execute, the models are executed inside Pega or externally by Google ML and the Amazon SageMaker platform.