Custom made notes to go through which boosts your confidence to crack Pega CPDC exam
Adaptive Models :
-> Applying only business rules enables to identify eligible propositions for a customer.
-> Improve acceptance rates by augumenting business rules with analytics.
-> ADM allows us to build self learning adaptive models.
-> Key capability of ADM is that it automatically detects changes in customer behavior and act on them in real time.
-> ADM is a closed loop system. It automates creation, deployment, monitoring process.
-> Used when customers behavior is volatile.
-> AM creates scoring model on fly and uses them for predictors.
-> AM can build channel specific models (differences in responses for outbound vs inbound offers for particular customer)
Examples :
-> real time detection of complex fraud patterns
-> predicting customer behavior after new product has been offered.
AM Cycle :
1> capture data from customer interaction
2> cration of reliable numeric intervals/sets of symbols using auto-grouping
3> assess inter-correlations in the data using predictor grouping
4> Establish uncorrelated view i.e. all relevant aspects to proposition by using predictor’s selection
5> using this adaptive scoring model for scoring customers
6> whenever new data is available, update scoring model
Diff b/w Adaptive and Predictive analysis :
-> PA requires historical data & human resources
-> Adaptive uses no historical data to calculate the likelihood of offer acceptance by capturing & analysing the real time data.
-> AM is created when a strategy using them is executed.
-> Predictors type : symobilc & Numeric
-> AM will inactive any predictors that are not useful.
-> Paramets passed to AM can also be used as predictos.
-> Define positive and negative outcomes. The values available to you for the positive and negative responses can be extracted from the Interaction History.
-> The ‘Performance threshold’ value is a limit that indicates that any predictor performing below this threshold will not be used to predict customer behavior.
Typically, the model creation will happen when you test a strategy that uses an Adaptive Model component.
Note that the same model configuration can be used for multiple physical models that are applicable to different propositions, channels or directions. It is strongly recommended that you set both the channel (.pyChannel) and the direction (.pyDirection) properties in your strategies before you use adaptive models.
AM Outputs :
1> Propensty (.pyPropensity) — 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%
2> 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.
3> Evidence — The number of responses used in the calculation of the Propensity.
Smooth Propensity :
-> Assumed starting propensity and starting evidence
-> Smoothe Propensity = @divide(SE, SE+ME+1.0 , 3 ) * SP + @divide(ME, SE+ME+1.0, 3) * pyPropensity
SE — Starting Evidence ; ME — Model Evidence ; SP — Starting Propensity
Coefficient of Concordance” (CoC) :
-> to measure the performance of predictors and models.
-> In the case of Adaptive models CoC measures how well a model is able to discriminate good cases from bad cases.
-> CoC is a value between 0.5 (random distribution) and 1 (perfect discrimination).
Trend Detection :
-> Trend detection compares the performance of multiple models.
-> To make this possible, the models that are triggered by the same proposition are configured with different memory sizes. his determines the time frame, in a number of cases, over which the performance is calculated.
-> The best way to think about the Memory setting is to compare it to learning speed. The lower the value, the quicker the model will learn. However, with a lower value, the model may pick up trends that don’t actually exist, as it is constantly changing its assessment of the situation.
-> This is why it is safer to adopt the size recommended by Pega, or at least champion/challenge your setting against a baseline model with a Memory setting of 0.
************************** PMML ************************
-> The Predictive Model Markup Language (PMML) is an XML-based language used to represent predictive models created as the result of a predictive modelling process.
-> This XML-based language is the de-facto standard to represent not only predictive and descriptive models, but also data transformations (data pre and post-processing).
PMML, like HTML is a Markup Language and as such is split into common components.
The latest PMML standard is 4.2.1 and Pega supports this version and earlier versions of PMML. Currently supported model types are:
· Cluster model
· General regression
· Mining (multiple/ensemble) model
· Neural Network
· k-Nearest Neighbors
· Naive Bayes
· Ruleset
· Regression
· Support Vector Machine
· Scorecard
· Tree
************************ Predictive Analytics Director PAD ************************
-> It enables data-savvy business people to create predictive models.
-> The final step (Model export) of the Pega modelling process Involves making the model available for use.
-> Once a predictive model is created, and configured using either a PMML model definition or a Pega model definition you must map the predictors defined in the model to the customer properties.
-> Finally you also select the desired output of the predictive model. The output of the model is mapped to the pxSegment strategy property when you reference the model in a decision strategy.
-> If you already use predictive models generated either externally or by using your own modelling tools you may consider using both models in your strategies. The Champion Challenger component selects predictive models randomly.
************************ Performance of Models ************************
-> Neither Pega’s predictive models nor PMML models output the runtime performance of an individual model.
-> To achieve this you need to create a feedback loop to compare the prediction of a model against the actual user response.
-> Pega Adaptive Models predict and learn in real time while continuously reporting on their performance.
-> In this pattern you use an adaptive model to monitor the performance of a predictive model.
If we have more than one predictive model we can then compare their respective prediction accuracy.
************************ Predictive Analysis ************************
-> 5 step wizard — PAD to create predictive models
1> Data Preparation : Select input data , define the behavior you want to predict
To import data from DB, specify the DB details in Database Settings
-> Source Selection
-> Sample Construction
-> Outcome Definition
2> Data analysis : prepare data, develop relationships between potential predictor groupings and the outcome predicted
-> Data Analysis
3> Model Development : Analyze how predictors work together & Creating predictive models using Regression and Decision Trees
-> Predictor Grouping
-> Hall of Fame
4> Model analysis : Analysis on performance. We check how customers are segmented according to predicted behavior
-> Score Comparison
-> Score Distribution
-> Class Distribution
5> Model Export : Generate model which includes definition of model output properties
-> Export
************************ Types of Models supported by PAD ************************
1> Scoring Models : prediction of binary behavior (examples : yes or no qustn)
2> Spectrum Models : prediction of continuous behavior (examples : likely)
***********************Decision Manager Portal **********************
In Business Sanbox — use to manage revisions of Decision artifacts, business rules, propositions, run test, run simulations to verify changes
In Production — monitor effectiveness of decision. Inspect interaction history using VBD and regular reports. Inspect details of adaptive models
DM Portal set up by System Architect
To set up :
1> System Architect creates an application overlay with create new application overlay wizard.
Application overlay has set of access groups associated with it and a set of business rules
2> In wizard, in first he gives name of application overlay and revision ruleset
3> Associate set of access groups: 3 default access groups and 2 access groups specific to prod
Used in Business Sandbox environment
-> Revision Mgr
-> Strategy Designer
-> Decision Architect
Other production related access groups
-> Administrator
-> Supervisor
4> System architect decide rules which can the business update.
5> The last step of the wizard lists all the artefacts which are created by the wizard. Once the wizard completes, the application overlay is listed in the landing page where all application overlays are displayed.
****** 5 different access roles and they have different sets of privileges **********
1> DecisionManager:RevisionManager
-> used in Business Sandbox environment
-> setup a new revision , create change requests, assign change requests to workbaskets/users, approve change requests , generate revision package
-> Privileges : pyTestFlows, pyManageRevisions, pyRunSimulations, pyManageVBD
2> DecisionManager:StrategyDesigner
-> used in Business Sandbox environment
-> working on change requests, change decision strategies, run simulations, test flows, submit changes for approval
-> Privileges: pyMonitorVBDSimulation, pyManageVBD, pyRunSimulations, pyWorkOnChangeSet, pyTestFlows, pyViewCalendar, pyMonitorReports, pyManageReports
3> DecisionManager:DecisionArchitect
-> used in Business Sandbox environment
-> applicable in all environments and it should be restricted to Decisioning experts only.
-> full access to system capabilities
4> DecisionManager:Supervisor
-> used in production environment
-> users need the ability to monitor the customer interactions and the performance of the adaptive models.
-> Have read only capabilities
-> Privileges: pyMonitorADMReports, pyUpdateADMReportingData, pyMonitorInteractionHistory, pyMonitorVBDActuals, pyMonitorReports
5> DecisionManager:Administrator
-> used in production environment
-> users need the ability to clean up adaptive monitoring data.
-> some responsibilities as they may significantly damage the production system
-> Privileges: pyMonitorADMReports, pyUpdateADMReportingData, pyMonitorInteractionHistory, pyMonitorVBDActuals, pyManageVBD, pyManageADMModels, pyTestFlows, pyMonitorReports, pyManageReports
maximize the profitabiliy of each customer To do this NBA bases each of its decision on 2 essential things
-> Customer needs right now
-> Business objectives for this interaction
centralized decision hub
4 key characteristics to satisfy customer needs
-> Consistent
-> Timely
-> Relevant
-> Contextual
4 key business objectives
-> Sales
-> Retention
-> Risk Mitigation
-> Service
Key features of pega DM :
-> Proposition Management — that can be offered to customer.
-> Decision Strategies — match propositions with customer, determine NBA for a customer during each interaction.
-> Interaction History — Pega DM’s long term memory.
-> VBD — 3-D graphical env for simulating and analysing the results of decision strategies and customer interactions.
-> ADM — automatic development and deployment of adaptive predictive models.These models learn data on-the-fly to predict future customer behavior.
-> PAD — environment to develop non-adaptive predictive models that use historical data to predict future.
Decisioning artifacts:
-> rule types — decision data, scorecard model,predictive model, adaptive model, interaction & strategy
-> landing pages — decisions LP, predictive analytics LP, Monitoring LP, Infrastructure LP
-> decision components — 7 groups
Import — external data, IH data, proposition to offer : data import, interaction history, proposition data
Business rules — eligibility requirements : DT, DTree, Map value, split
Decision Analytics — make predictions of customer i.e. calculate likelihood : AM, PM,SM
Enrichment — calculations : data join, decision data, set property
Arbitration — filter out propositions : contact policy, filter, geofence filter, prioritize, segment filter
Selection — ranking propositions : Champion challenger, exclusion, switch
Aggregation — calculation of aggregate numbers : group by, iteration, financial calculation
-> flow shapes — decision(decision table, decision tree,predictive & scorecard model) and run interaction
6 dimensions available in VBD :
actions,channels,applications,operators,outcome,times
y-axis : propositions
x-axis : outcomes
Adaptive models use customer properties as predictors
Applications can enrich the number of components available to the strategy designer. ?
For example, Pega Marketing makes three additional components available: Contact Policy, Geofence, and Segment.
As soon as we bring an adaptive model component into a strategy, the data collection begins.
Adaptive Decision Manager analyzes the data it has collected, applies the automated modeling process to it, and the end result is a predictive model.
Once the proposition is offered, and the customer’s response is captured, the new evidence is given to Adaptive Decision Manager, which uses it to refine the existing models. When new models are ready, Adaptive Decision Manager deploys them automatically into the Pega Decision Management decision engine.