Evaluating Your Prediction

Once your prediction has a Ready status, it's ready to be reviewed.

# In this article

# View your prediction

Navigate to Insights > Predictive Goals and select the goal you want to view.

On the Goal Details page, you'll see information about how likely users are to convert on your goal from the last refresh date through the period of time you specified (for example, 1 month).

Some of the information you see will look familiar—you defined the goal criteria, the contact property name, and a description when you created the goal.

The rest of the information is new, so let's take a look at what's most helpful when evaluating predictions.

# Predictive strength

Predictive strength is a quality gauge that tells you how reliable a prediction is likely to be.

Value	What it means
Strong	A prediction's ready for use, with expected high accuracy. Changing the goal criteria is unlikely to improve the quality of the prediction.
Satisfactory	Also a reliable prediction. You could try changing the criteria, but you don't have to in order achieve reliable results.
Weak	Not a reliable prediction. Update the predictive goal criteria, check whether associated events and properties (user profile fields) have enough data, or wait for the next refresh cycle to see if the predictive strength improves.

# Review Explainable AI prediction insights

Exploring details of the events and properties involved in your prediction can be incredibly insightful. To access this information, from Understand your prediction, click Explore. You'll see a panel that looks something like this:

This view shows the number of events and properties that Predictive Goals evaluated when the prediction was generated (in this case, 120). It also shows a breakdown of the statistically significant contributors.

Let's consider an example predictive goal that looks for users who are likely to be highly engaged with your brand. Based on this sample result, you can see that Predictive Goals found that events such as making a purchase, viewing your blog, or starting a trial are good indicators of user engagement.

In addition, you can view the events and properties that are likely to increase or decrease the likelihood of goal outcome. Using the previous example (increasing engagement with your brand), you'd want more users to create accounts, make purchases, and view your blog as reflected by the AccountCreated and BlogViewed custom events, and by the Purchase event.

And, you'd want fewer instances of users canceling orders and unsubscribing, as reflected by the CancelledOrders and Unsubscribed custom events.

# Segment-specific details

In the Prediction area, you'll find a histogram chart with Percentile on the x-axis and Probability on the y-axis.

• To focus on a portion of your users, select a range of Percentile values. Use the radio buttons to quickly select the top and bottom 10% of your users, or select a custom range by entering specific values (or using the sliders).

• To understand the likelihood that the selected users will convert on your goal, examine their Probability values. These values vary based on the results of your prediction.

As you change the selected range of users, the chart's details refresh. Interactive values include:

• The number of users included in your selection and how likely they are to convert on the goal compared to the average of all users.

• The multiplier of how much more likely to convert the selected users are compared to the average of all users.

• The approximate number of conversions you might expect on all or part of the goal criteria. This value is calculated by multiplying the average probability by the total number of selected users.

  

For considerations involved in choosing a user segment, see Creating user segments and personalized messaging.

# Next steps

Now, learn how to create segmentation from your prediction. See Creating user segments from predictive goals.