Manage Next Best Actions

When you open a next best offer prediction, the details page helps you understand how the prediction is performing and how to use it in your campaigns. It summarizes what the prediction is measuring, shows key quality and refresh metrics, and breaks down how different ranges of users are expected to behave. You can use this view to review model strength, compare potential target ranges by reach, CVR, and lift, and save the range you want as a new audience for activation.

Prediction Recap

The Prediction Recap gives you a quick summary which actions or offers the prediction is choosing between, the business objective it is trying to drive, and the users that are eligible to receive an offer, including the current estimated number of users. Use the Prediction Recap to confirm that you are looking at the right combination of offers, business goal, and users before you review results or use the prediction to create new audiences.

Prediction Results

The Predictions Results provides everything you need to understand how reliable your prediction is and how to put it to work in your campaigns. You see key metadata about prediction strength, when it was last refreshed, how often it updates, and its current status.

The Target Range selector lists the estimated reach, predicted conversion rate (CVR), and predicted lift for different ranges of users. The Likelihood Percentiles chart shows how users are distributed from least likely to most likely to act on their best fitting offer. Together, these pieces help you balance reach and efficiency, so you can decide which users to target, and how narrow or broad your audience should be.

Prediction Strength

Prediction strength reflects how accurately the prediction can identify users who are likely to perform the target behavior. It is based on the quality of your data, the strength of behavioral patterns, and how well the prediction’s training process performed.

• Weak: The model strength is low. Behavioral signals are limited or inconsistent, so results should be interpreted with caution and not used for campaign targeting.
• Moderate: The model strength is acceptable. There is a meaningful behavioral pattern, but still some uncertainty. You should use these results alongside other insights when planning a campaign.
• Strong: The model strength is high. Behavioral patterns are clear and consistent, so results are very reliable and should be used to inform campaign strategy.

If a prediction’s strength is lower than expected, verify that you are sending enough historical events to mParticle and that the target behavior occurs frequently enough to provide meaningful training data.

Last Refresh

Shows the last time the prediction was recalculated, according to the prediction’s refresh frequency.

Refresh Frequency

The refresh frequency is how often your prediction is recalculated using the latest available data. It tells you how frequently mParticle updates the prediction scores for your users, so you know how current the results are before using them to create and activate audiences.

Update a prediction’s refresh frequency

To update a prediction’s refresh frequency from the details page:

1. Click the three-dot action icon from the prediction’s detail view (or from the Predictive Attributes page).
2. Click Update frequency.
3. Use the left-hand dropdown to select one of the following intervals: Weekly, Monthly, or On demand.

4. Use the right-hand dropdown to specify when or how often (within the selected interval) the prediction should refresh.

   • For example, if you select Weekly, choose which day of the week to refresh.
   • If you select On demand, the prediction will only refresh when you click Run Now from the action menu.
5. Click Save.

Run now

You can recalculate a prediction manually by clicking Run Now from the prediction’s action menu, either on the detail view or from the Predictive Attributes page.

Running a prediction on demand updates its results immediately using the most recent data available.

Status

The status shows the current state of each prediction’s pipeline.

• Calculating: mParticle is in the process of generating predictions. Predictions that are currently being calculated cannot be used in audiences.
• Active: The prediction has been calculated and is ready to be used in an audience.
• Inactive: The prediction’s pipeline has not been run within the last 30 days.
• Failed: mParticle could not complete the prediction calculation for your users. This often happens when the number of eligible users is too small or there is an issue with the underlying data. To diagnose and resolve the issue, contact Support.

Target Range Definition

The Target Range section helps you explore different slices of your scored users and choose the one you want to turn into an audience.

You can select all users or define a custom range, then compare how each option will perform before you commit it to a campaign. For the selected range, mParticle shows:

• Estimated Reach: How many users fall within the selected range.
• Predicted CVR: The predicted conversion rate for sending that audience their best offering.
• Predicted Lift over Random: How much higher the predicted CVR is over sending each user a random offer.

Click the dropdown menu to switch between selecting all scored users or creating a custom range:

• Include all scored users (0.00 to 1.00): Allows you to target all evaluated users, matching each user with their best offer. Select this option when you want to maximize reach.
• Include likely converts (Custom top percentiles): Allows you to target a custom percentage of the most likely converters. After selecting this option, use the arrows or enter the percentage you want to target in the text box that appears to the right of the dropdown.

When you enter a custom target range, it will update the values in the table on the left and highlight the range in purple in the Likelihood Percentiles Chart on the right, giving you a visual presentation of the predicted performance.

Save a target range as a new audience

Once you’ve identified a user range you want to target, you can create an audience directly from the predictions details page:

1. From the prediction details page, once you’ve selected your target range, click Save as New Audience.
2. The Audience Builder will open with the criteria selecting your targeted users prepopulated.
3. Save and activate the audience to use it in downstream campaigns in the same way you activate other audiences.

Predictive audiences automatically stay up to date as your predictions refresh each week, helping you maintain targeting accuracy across changing customer behavior.

Likelihood Percentiles Chart

The Likelihood Percentiles chart shows how your scored users are distributed from least likely to most likely to perform the target action, and how the predicted conversion rate changes across that spectrum.

Users are grouped into 10 equally sized percentile ranges (deciles) along the x-axis, and the number of users is shown on the y-axis. You can use this chart to balance reach and efficiency.

The chart includes several visual elements:

• The colored areas of the chart represent each offer included in the prediction.
• Selected range: When you select a custom target range, a purple shaded area highlights the percentile range in the Likelihood Percentiles chart. This makes it easy to see where your chosen audience sits on the likelihood spectrum.
• Average CVR line: The black line represents the average predicted conversion rate across all scored users.

Troubleshooting low-quality or inactive predictions

If your prediction shows a Fair or Poor quality rating, or its status becomes Failed, check the following:

• Verify that enough historical and recent event data is being ingested for both the target behavior and evaluated offers.
• Confirm that your prediction’s audience segment isn’t too small or overly filtered.
• Review the offer distribution chart to ensure one offer isn’t dominating, which can reduce training effectiveness.
• If data sufficiency or variety is low, consider expanding the prediction’s timeframe or audience scope.

By reviewing these elements, you can improve prediction stability and maintain accurate, actionable recommendations.