PEGA Decisioning - Predicting Customer Behavior Using Real-time Data

·         Applying adaptive analytics to your strategies will enable them to detect changes in customer behavior as they occur and act on them immediately

·         Adaptive Decision Manager is a closed-loop system that automates the model creation, deployment, and monitoring process. It can manage a large number of models without human intervention.

·         ADM can capture and analyzing response data in real-time. cases where data is available for offline modeling, predictive models can be used as an alternative, or in conjunction.

·         ADM models to be created under Customer class, Decisioining, Adaptive Model

·         Any Single value Property from Customer class can be a predictor. It can be added under Predictors tab. It can be either symbolic/Numeric. PEGA ADM will automatically remove the unused predictors

·         Define Positive & Negative Outcomes under Outcome Tab

·         On the Settings tab of the Adaptive Model rule instance, you can define values that will fine tune the behavior of a model as a whole. For example, the ‘Performance threshold’ value is a limit that indicates that any predictor performing below this threshold will not be used to predict customer behavior.

·         When you configure an adaptive model record no models are created. The models are created on-the-fly, once there is demand for a given model. When it evaluates the Adaptive Model component, it requests the propensity for each single proposition or channel combination that is dictated by the component inputs. If the models do not exist, they are created at this time

·         It is strongly recommended that you set both the channel (.pyChannel) and the direction (.pyDirection) properties in your strategies before you use adaptive models. The channel and direction settings enable you to make decisions based on customer behavior in a specific channel.

·         Adaptive Model Outputs - Propensity, Performance, and Evidence

Propensity – This is the predicted likelihood of positive behavior. Such as, the likelihood of a customer accepting an offer.  The propensity for every proposition starts at 0.5 or 50% (the same as a flip of a coin) because in the beginning the model has no response behavior on which to base its predictions.

Performance: This is how well the model is able to differentiate between positive and negative behavior. Again, the initial value for each model is 0.5, with 1.0 being perfect performance. Therefore, the performance value should be somewhere between 0.5 and 1.0. Performance is generally used to differentiate between two models relating to the same proposition.

Evidence: – The number of responses used in the calculation of the Propensity.

In strategies, model propensity is automatically mapped to the strategy property called .pyPropensity. 

There is no automatic mapping for the Performance or Evidence outputs. But they can be manually mapped to any of the strategy properties under the Output mapping tab (ModelEvidence, ModelPerformance).

Smoothened Propensity: To reduce the ADM Propensity error in early days.

(@divide(.StartingEvidence, (.StartingEvidence + .ModelEvidence+1.0),3) * .StartingPropensity) +

 (@divide(.ModelEvidence, (.StartingEvidence + .ModelEvidence+1.0), 3)*.pyPropensity)

Where: Starting Evidence and Starting Propensity are proposition properties representing the assumed values for evidence and propensity.

Pega Decision Management uses “Coefficient of Concordance” (CoC) to measure the performance of predictors and models.

 

 

 

Trend Deduction:

·         How to measure the performance of the new Proposition using Adaptive, Trend Deduction.

·         Define the required Predictors, Outcome (keydown will prompt IH Data options) and memory settings (Run analysis after 500 records)

How to define More than one ADM for the same propositions?

·         Define the 2 diff ADMS using different Memory settings

·         Add 2 ADMs and add GroupBy Proposition (pxIdentifier) with the highest performance. Now, we should able to select the highest performance Proposition.

 

All ADM models Management options can be found under ADM Management.

o    Clear Model - Clears the model responses

o    Delete Model - Removes the physical model

o    View Model Properties - see the settings of an Individual Model parameters

o    Upload Responses will allow you to import the Historical data. upload the CSV file and map the outcome.

Monitoring the Adaptive Models:

·         DS->Decisioining->Monitoring->Adaptive Models Reporting

·         You can analyze the Behavior report under Reports to analyze the Active/Inactive predictors. Also the Classifier allows you to analyze the propensity.

·         Performance Model allows you to analyze the performance of the ADM

·         The Predictors overview will enable you to analyst the various predictors defined

The implementation of the Next-Best-Action mechanism is a staged process with each stage refining the proposition selection process.

PEGA Decisioning - Avoiding Loan Default with Predictive Analytics

For predictive analytics algorithms to work we must have access to historical data which exhibits known customer behavior. We must also know what problem we are trying to solve in our case probability to default.

These fields are called predictors and are combined into a predictive model which you can use in your business processes.

Predective Model involves 5 Steps:

Data preparation, Data analysis, Model development, Model analysis, Model export

Predictive analytics director supports two types of models:

·         Scoring Models - for the prediction of binary behavior

·         Spectrum Models - for the prediction of continuous behavior

Scoring Models:

The value calculated by the model, known as the score, places a case on a numerical scale. High scores are associated with better business (good behavior) and low scores are associated with worse (bad behavior). Typically, the range of scores is broken into intervals of increasing likelihood of one of the two types of behavior. Scoring models require behavior to be classified into two distinct forms like positive and negative. Classic examples of such behavior are:

·         Responding to a mailing or not

·         Repaying loans or going into arrears

Spectrum Models:

Spectrum models extend the ideas of scoring models to the prediction of continuous behavior

·         Likely purchase value of responders to a direct mail campaign

·         The likely eventual write-off of cases recently falling into arrears

Model Template:

In the Predictive Analytics Director portal we have defined a number of model templates which you can use. These include Risk, Retention, Recruitment, and Recommendation.

 

Select the appropriate Model Template and start working on the Project

1) Model Creation:

·         Select source as CSV file or DB

·         Set the Sampling size using % & tot. cases. Define the properties to be used and the type of the property

·         You identify the field which you are trying to predict. In this example, the field is behavior. The field exhibits binary characteristics (N/Y) so the scoring model is the most appropriate.

·         Now define Good & Bad behavior under Outcome definition.

2) Data Analysis:

·         Predictive Analytics Director facilitates automatic discovery of correlation patterns of individual predictors and their ability to predict the outcome. Any unique identifiers will not be a valid predictors. customer ID appears to be a reasonably well performing predictor. However, since we know that customer ID is a random number or a member of a sequence, it cannot have any impact on the good or bad behavior. Thus we remove IDs from the candidate list for predictors. Any property which is common to use to differentiate the good and bad behaviors can go as a predictor

·         So data analysis involved in defining the properties & Binning (Split the data into Buckets) and then grouping by predictors.

3) Model Development:

·         Predictive Analytics Director provides a rich model factory supporting industry standard models such a regression and decision tree models.

·         The system automatically creates two models: the Regression and the Decision Tree-CHAID model. At this stage, you can create additional models if required.

·         We can group the predictors into Group,. Which means either Predictor A/B/C outcome will be considered. Eg. HouseOwner/Rented etc.

·         Also look at the scorecard representation model to check the scores of the predictor and define the weight of the predictor from 0-1000

4) Model Analysis:

·         Model Analysis is to enable you to create a shortlist of models and then select the best model for your use case. At this stage, you also group the scores into statistically significant set of score bands; firstly, let’s examine the relative performance of individual models.

·         A very important aspect of each model is its performance, i.e. how good is a model or a given predictor in predicting the required behavior. We use a term “Coefficient of Concordance” or (CoC) for the measure of the performance of predictors and models. You could describe CoC as a measure of how good the model is in discriminating between good cases from bad cases. The value of CoC ranges between 50%: a random distribution, and 100%: the perfect discrimination.

·         Analyze the Models created and select the models where how good the model is in discriminating between good cases from bad cases.

·         There are following steps in Model analysis: Score Comparison, Score Distribution and Class Comparison.

5) Model Export:

·         In the final stage, you can produce the reports about the model and export the model into a model file or an instance of the Predictive Model rule. Check the customer properties and the predictors created in the model under input mapping.

·         The report contains:

1.     A project summary

2.     A visualization of the whole Decision Tree

3.     The sensitivity of the model for each of the input fields

4.     Model segmentation

5.     Detailed insight in the analysis, grouping, and validation of each of the attributes

6.     Date when the model was developed and by

PEGA Decisioning - Avoiding Redundant Product Offers to Customers

How to make the Offer recommended to the Customer is Personalized? One basic eligibility check is by avoiding Redundant Product Offers to Customers

• If customer already have the Products, don't offer them.
• We can extend the Customer DataModel by defining the 1-1 OR 1-n relationship of the customer owned Products/services. For example, a list of purchases made, a list of accounts owned, a list of family members, and a list of services owned.
• You can define the Relationship between Customer class & Product class
• You can build the Custom DataFlow to populate the customer Profile information composing other Data properties
• Nested Data Type also possible. We need to discuss and agree with the Data Model & compose the same with the respective class object
• When u use the Data Join, the target mapping always be the SR class properties and the source can be customer class properties OR another SR class properties
• Data-Join component allows you to join 2 sets of data and map to the SR class properties
• Filter component Filter can be plain Filter condition (Takes Customer class properties & SR class properties) OR it can be a Proposition Filter.
• @and(booleanA, booleanB) validates to TRUE if both the conditions met
• When u connects with the newly created DB table, create the Table and respective classes through DB Class mapping wizard.
• Go to Data Model -> Classes & Properties -> Database Class Mappings landing page. Click New External Database Table Class Mapping.
• Use DataImport component to Import the Data from External DB Table contents.

To avoid the same product recommendation which customers own already, from PEGA we need to define the data relationship using DataFlow. For example, to cater to a use case where a customer owns multiple products and details of all products are available in a Product master table. So, there is a one-to-one relationship between the ProductHoldings and Product entities.

Now to implement the business case discussed (Filter the products already owned by the customer), you must filter out the products already owned by the customer. For this, you use a strategy pattern that is composed of three components: Data Import, Data Join and Filter, we can filter out the products which customer is holding already.  DataImport+DataJoin+Filter is the Strategy design pattern used to achieve.

 

Update the Strategy to include a component that retrieves Interaction History data for previous phone offers. We then filter out phone offers for customers that have accepted a phone offer in the past. We can of course make the filtering process much more sophisticated by considering phone upgrades, etc. But for the sake of our demonstration, we will keep the strategy simple; anyone who has been made a phone offer within the last 120 days will not be made another phone offer of any kind. IH+GroupBy+Filter is the Strategy pattern used to prevent Redundant offer recommendation in PEGA Strategy logic.

Happy Learning!