Basically, Pega adaptive model is a self-learning AI technic which doesn`t need any historical data. In a Pega strategy, business rules can be used to choose Next-Best-Action, but it doesn`t give appropriate offers all the time. Pega`s adaptive model fix it.  Before getting any historical data, it works randomly and collects data. After that, it detects customer behavior, and offer most suitable next-best-action for a customer. So how is this work? Let`s examine together.

Firstly, we need a strategy but to create a strategy, we should create a business issue and a group. Business issue is the main subject about what you want to offer to a customer. Group is the specific thing into the business issue. To creating them,

 Dev Studio>Configure>Decisioning>Decisions>Proposition Management>Hierarchy>Add New Business Issue/Add New Group



Now we can create a strategy,


Select the business issue and group you just created and create the strategy.


Now you have a strategy to create Pega Adaptive Model.

Click + (plus) icon>Decision Analytics>Adaptive Model to add Adaptive Model shape.

Create an adaptive model inside the adaptive model shape. In the adaptive model, there is a predictors tab. In this tab, we can specify predictors using the model. Context tab covers the areas which will define after the model completed. In the outcomes tab, we choose which outcome impact as negative or positive affect. Also, in monitoring tab, we can monitor the model`s context.


After we create adaptive model, now we can see the model`s context and predictors in adaptive model shape. This shape gets information from source component and it automatically has output mapping to .pyPropensity. If you add more, use output mapping tab. 


Now we have an adaptive model shape, so what we are going to do about it? We will connect it with other properties that we need. Adaptive model predicts customer behavior using propensity. Propensity is the possibility of customer acceptance. But we don`t have any historical data and early days of adaptive model can be unstable. So, we use smoothed propensity to balance it. Recommended formula of smoothed propensity is:

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

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

Now we have a basic adaptive model, you can improve your strategy for your business needs.

I appreciate to read my post.