When you create email channel, a predictive model is automatically generated behind the scenes. This model is training-ready. In demo video - https://www1.pega.com/insights/resources/intelligent-virtual-assistant-email, at 0:46 CSR creates a business case named 'New user request'. This tells the underlying model that for a given customer email, identified category was 'New user request' and actual result is also 'New user request'. If CSR had selected anything from 'Other cases' dropdown like 'Add nominee', model would be told that for a given customer email, identified category was 'New user request' but actually it is 'Add nominee'. This feedback/training data stays in database as a Training record and is available to view/edit on 'Training data' tab of corresponding email channel. Here, a data scientist can assign correct category for given customer email and add training records to model. When number of training records are substantially high the model can be built from Analytics center and that is how a model will learn to identify categories correctly the next time or return results with greater confidence.

When you create email channel, a predictive model is automatically generated behind the scenes. This model is training-ready. In demo video - https://www1.pega.com/insights/resources/intelligent-virtual-assistant-email, at 0:46 CSR creates a business case named 'New user request'. This tells the underlying model that for a given customer email, identified category was 'New user request' and actual result is also 'New user request'. If CSR had selected anything from 'Other cases' dropdown like 'Add nominee', model would be told that for a given customer email, identified category was 'New user request' but actually it is 'Add nominee'. This feedback/training data stays in database as a Training record and is available to view/edit on 'Training data' tab of corresponding email channel. Here, a data scientist can assign correct category for given customer email and add training records to model. When number of training records are substantially high the model can be built from Analytics center and that is how a model will learn to identify categories correctly the next time or return results with greater confidence.