Adaptive analytics is a form of predictive analytics that augments decision strategies with self-learning adaptive models to improve predictions about how customer interests and needs change. The Adaptive Decision Manager (ADM) service automatically detects changes in real-time and acts on them to adapt to customer interactions.
Adaptive decisioning captures and analyzes data to deliver predictions in situations where historical information is not available and continuously increases the accuracy of its decisions by learning from each response to a proposition. For example, if a customer is offered a product and accepts it, the likelihood that customers with a similar profile will accept that offer also increases slightly. More precisely, the mathematical expressions of these probabilities in the model are automatically updated after each positive or negative response.
Adaptive models do not need historical data to develop a predictive model that has a good performance because adaptive models can calculate who is likely to accept or reject an offer using no historical information. Adaptive models do this by capturing and analyzing response data in real-time. This is particularly useful in situations where the customer behavior changes often. A typical use case is the real-time detection of complex fraud patterns or predicting customer behavior following the introduction of a new offering.
Adaptive model in a strategy
Configure the ADM Service
Add decision data nodes to enable adaptive analytics capabilities in your application. Configure the Adaptive Decision Manager (ADM) service that enables each adaptive model to develop knowledge about customer preferences through their interactions in real time. For more information see:
Create adaptive models
Define adaptive model configuration through an instance of the Adaptive Model rule. The models themselves are created automatically when the strategy that references the Adaptive Model rule is run. Create Adaptive Model rules in the Customer class of your application.
When defining your adaptive models, you must specify the properties that the adaptive models use as predictors to learn from. You can use any single value property as a predictor. Predictors can be symbolic or numeric, as shown in the following example:
Defining predictors in an Adaptive Model rule
Finally, you identify the responses that indicate positive or negative behavior. You can use more than one value as an example of a positive or negative behavior. Different applications may use different words to identify positive or negative behavior, for example, Positive-Accepted, Negative-Rejected, Purchased, Rejected, and so on.
Defining adaptive model outcomes
For more information, see:
Manage adaptive models in your application
Manage adaptive models that are specific to your application on the Adaptive Models Management landing page or in the Decision Analytics work area. You can view last responses to make sure that the model learning process is in motion, clear models to remove adaptive statistics associated with them and delete the models that you no longer need. For more information see:
Pretrain adaptive models
When you first start using adaptive models, they do not have information on which to base their predictions. The predicted likelihood of the positive behavior, called propensity, is 50 percent. The propensity will change with every response that the models get. However, collecting responses takes time, and you might have some customer data that you can upload to accelerate the learning process. For more information, see:
Get smarter about how you manage your adaptive models. Information that you can obtain by monitoring the performance of each adaptive model instance in your application is important for business users (business scientists, data scientists, and business intelligence specialists) and Strategy Designers. This information can influence changes to the decision strategy and adaptive models. For more information, see: