Predictive Scoring

[This topic is pre-release documentation and is subject to change.]

What is Predictive Scoring?

Build a predictive scoring model to enable business users of all skill levels to apply artificial intelligence to predict a range of business outcomes. Based on these predicted outcomes, users can grade customers to efficiently plan targeted follow-ups.

Use Predictive Scoring to answer questions like:

  • Which of my leads should I focus on?
  • Which cases can escalate?
  • Which opportunities will convert this period?
  • Which of my accounts are at risk of churn?

Customer Insights Predictive Scoring has the following benefits:

  • Business users can predict ANY business outcome
  • Automatic discovery of the best predictive signals
  • Continuous scoring of incoming data
  • Interpretable predictive factors
  • Grading for business process management

Create a new prediction

  1. Open your Customer Insights Customer 360 application.

  2. Select Show Menu .

  3. Select All Options > Predictive Analytics.

  4. Select Add and fill in the values.

    Prediction Details

    Item Description
    Prediction Name The name of the prediction.
    Description A description of the prediction.
    Score Label Describes how the score will appear in all the result visualizations.

    Define Outcome to Predict

    Item Description
    Profile Select a profile to whose outcome you want to predict.
    Positive Outcome Values You will need to specify search criteria based on your selection.
    Negative Outcome Values You will need to specify search criteria based on your selection.

    Apply Prediction

    Item Description
    Prediction can be added to new or existing attributes of the profile Default: Add new attributes to target profile.
    Score
    Reason
    Grade
  5. Select Create to save your new prediction. Your model will be trained and validated by analyzing historical data based on the profile you've selected. The process will take a few minutes before being available for your analysis.

Sample Predictive Scoring settings

The following values were used to create a sample prediction of account retention based on historical account activity data.

Prediction Details

Item Value
Prediction Name AccountRetentionPrediction
Description Accounts retention prediction
Score Label Retention

Define Outcome to Predict

Item Value
Profile Account
Positive Outcome Values Status Is Active - these are accounts which are still active
Negative Outcome Values Status is InActive - these are accounts that have churned

Apply Prediction

Item Value
Prediction can be added to new or existing attributes of the profile Add new attributes to target profile.
Score PredictionName_Score
Reason PredictionName_Reason
Grade PredictionName_Grade

The new policy model builds with the above values and then validates.

Once completed, you can review the prediction results.

In this example, note the following:

  • The red and green areas of the chart show the separation between the accounts successfully retained and those that have churned. The more successful the model prediction the more these areas are separated out.
  • The expected model accuracy is found in the 76% and 93% outcome accuracy.
  • The baseline account retention is 71%. This view also shows the top 5 predictors for the retention score.
  • Grading was set so that a score of 68 or higher received an "A" for likelihood of rention and a "B" if less than 68.

Once the prediction is available you can look at the overall performance of multiple predictions.

View the overall performance of a prediction once it has been running for a while.

In this example, note the following:

  • In Prediction Performance, you can see the retention score versus the actual percentage of retained accounts. This is the productivity gain resulting from this prediction model.
  • Retention by Grade shows the actual retention for different grades. Green shows accounts successfully retained. Red shows accounts lost.
  • Retention Rate by Grade shows predicted versus actual retention rates.
  • Top 5 Account Profiles are the accounts most like to be retained.
  • Bottom 5 Account Profiles are the accounts most like to be lost.

Prediction scoring can be added to your Customer 360 view.