Capturing customer responses

A Make Decision data flow uses adaptive models to determine the next best action. A Capture Response data flow matches the customer behavior towards the action with the initial decision. Together, you can use these data flows to demonstrate that adaptive models learn from customer responses.

This video describes how adaptive models learn from customer interactions by running the ‘Make Decision’ and ‘Capture Response’ processes.

Consider this scenario: when a customer logs in to the U+Bank website, a personalized banner promoting a credit card offer should be displayed. Behind the website, the Customer Decision Hub uses an adaptive model to optimize the click-through rate for each banner.

Let’s see what happens inside the Customer Decision Hub in this scenario.

When a customer logs in to the U+Bank website, a decision on which banner to show them is requested from the Customer Decision Hub. This triggers what we call a ‘Make Decision’ process. The Make Decision process is implemented using a data pipeline called a data flow.

The first step in the data flow is to retrieve the input data required to make a decision. A subset of this input data is used as predictor values during the adaptive model learning process. Once the data is loaded, a decision strategy is executed to determine the credit card banner that should be shown to the customer on the web site. This strategy uses an adaptive model to predict the banner the customer is most likely to click on.

The Customer Decision Hub stores a transient snapshot of all predictor data as well as the new decision made for the customer and presents the banner on the website.

The ‘maximum response waiting time’ is the time period in which the system is waiting for a customer response. In a Web use case, the ‘maximum response waiting time’ in which a customer can click on the banner to generate a positive response is set to 30 minutes. However, when the objective is to detect responses to an outbound email, the ‘maximum response waiting time’ is usually set to a couple of days, because email is not a real-time channel.

So, let’s see what happens when a customer is interested in the offer and clicks on the banner within 30 minutes of its presentation. The ‘Capture Response’ data flow is triggered. It uses the snapshot containing the decision and the customer’s response as input.

This data flow contains a response strategy that translates the customer behavior into a positive outcome. The decision-response pair is sent to the Adaptive Decision Manager and used for the adaptive model learning process. The updated adaptive model will reflect a slightly higher likelihood of click-through by customers with similar properties. Likewise, if the customer does not click on the banner within 30 minutes, the updated model will predict a slightly lower likelihood of click-through for a similar customer. This is how the Customer Decision Hub learns from every customer response.

You have reached the end of this video. What did it show you?

• How a ‘Make Decision’ data flow uses adaptive models to determine the credit card banner that is most likely to get a positive response.
• How the ‘maximum response waiting time’ determines the timeframe in which positive responses to an offer are recorded.
• How a ‘Capture Response’ data flow is used to match customer behavior with the initial decision.
• How the adaptive model learns from the customer interaction.