This article supports Audience Studio - Legacy.

You can apply Treasure Data’s machine-learning feature, Predictive Scoring, to your customer or account segment. Use Predictive Scoring to assess behaviors and predict future behavior.

Depending on the configuration of the segments, you can solve a wide variety of wants that predict promising profiles. As long as a subset of profiles can be represented as a batch segment in a master segment, Treasure Data can predict customers or accounts who are likely (or, if needed, unlikely) to be in the segment.

As a consequence of scoring, you can see the distribution of scores and corresponding metrics on the dashboard. Additionally, activation for likely profiles can be seamlessly implemented from the dashboard. From the screen, you can see the customers divided by a score indicating the likelihood that the customer will churn. You can also see a chart with the distribution of users in the segment by their likelihood of churn. You can also use the page to create a new segment.


Scroll down the page to see attributes that contributed to the prediction, accuracy estimates, and values for important attributes:

Predictive Scoring Use Cases

This section summarizes possible use cases for the Predictive Scoring feature.

Predict Future Customer Churn

  • Population: All customers in a master segment

  • Positive samples: Already churned customers

  • Scoring target: Not churned customers


Predict Future Conversion in US

  • Population: People in US

  • Positive samples: Converted customers

  • Scoring target: Not converted yet customers in US

Predict Customer Purchase in 2018 Based on 2017 History

  • Population: People who accessed in 2017

  • Positive samples: Purchase log in 2017

  • Scoring target: People who accessed in 2018

Predictive Scoring Terms

Term

Description

population

a set of profiles defined as a master segment or as a segment. The set is used to build the predictive model; the characteristics of customers or accounts are extracted and patterns are learned, and the predictions are based on the patterns learned from the segment.

positive samples

indicate your definition of conversion.

For example, if you want a customer churn prediction, then this segment is a set of customers who were already churned.

scoring target

specifies profiles you are interested in; the prediction is done only for the profiles in this segment.

overfitting

a modeling error that occurs when a function is too closely fit a limited set of data points. Data is overfitted if AUC is very low or exceptionally. Sometimes this situation is referred to as overtraining. Overfitting occurs when the prediction model memorizes the data points in their entirety rather than by learning patterns.



  • No labels