Skip to main content

What to do if my prediction is stuck in 'Queued' state?

Updated

This article explains what the “Queued” status means for Predictions and what to do if your prediction appears to be stuck there.

It covers:

  • What “Queued” means in the prediction lifecycle
  • How long can a prediction normally stay in “Queued”
  • Checks you can perform yourself in the UI and configuration
  • When to restart the prediction
  • When to contact Bloomreach Support and what information to provide

What does “Queued” mean?

After you configure and start a prediction in Analytics -> Predictions, Bloomreach runs a machine learning job to train the model. During this process, the prediction goes through several statuses on the Results tab:

  1. Queued – The prediction has been submitted and accepted. It is waiting in an internal queue and hasn't started training yet.
  2. Processing / Training – Data is being prepared, and the model is being trained.
  3. Completed / Ready – The model finished successfully, and the prediction attribute is ready to use.
  4. Failed / Aborted / Stopped – The model stopped due to an error, was aborted by the system, or was stopped manually.

Immediately after you click Start, it is expected to see Queued as the first status on the Results tab:

 

How long is it normal to stay in “Queued”?

The time a prediction spends in Queued status depends on your data size and configuration, but you can use these general guidelines:

  • Typical behavior
    • “Queued” appears immediately after you click Start.
    • The prediction usually moves from Queued into processing/training within a few minutes.
    • A standard prediction (for example purchase, churn, optimal send time) usually completes training within 10 minutes to 2 hours.
    • For very large datasets or complex custom predictions, a few hours can still be normal.
    • In some cases, the full training process, including time spent in the queue, can take up to 24 hours.
  • Potentially abnormal behavior
    • The prediction remains in Queued for more than ~2 hours and never transitions to any processing or training status.
    • The prediction remains in Queued even after you stop and restart it.
    • The prediction has been in Queued continuously for more than 24 hours.

If your prediction has been in Queued for > 2 hours, it is worth performing the checks below. If it remains Queued for > 24 hours even after a restart, something is likely wrong, and you should contact Bloomreach Support. 
You can find information on how to contact support here.

 

Step 1 – Confirm that the prediction is really “stuck”

  1. Open the prediction Results tab

    Go to:

    • Analyses -> Predictions
    • Click your prediction.
    • Open the Results tab.

    Check the Status:

    • Immediately after starting, seeing Queued is normal.

    • If the status never changes from Queued to a processing/training status for a long period, the prediction may be stuck.

       

  2. Give it enough time

    Before assuming the prediction is stuck:

    • Wait at least 30–60 minutes for small/typical datasets.
    • For very large datasets or complex custom predictions, it is sensible to wait up to a few hours.
    • If the prediction has been Queued for more than 2 hours with no visible progress, continue with the checks below.

Step 2 – Check basic data and configuration

Next, verify that the prediction configuration and underlying data are reasonable. Several issues can make training impossible or extremely slow.

2.1 Check data availability and volume

In general, predictions work best when:

  • There is enough historical data for the event you are predicting (purchases, churn, email opens, and so on).
  • There are sufficient eligible customers to train on.

Recommended checks:

  1. Is your project new or sparsely populated?
    If you only recently started tracking events or have very low traffic, the model may not have enough data to train effectively.
  2. Rough data baseline
    • As a rule of thumb, it helps to have:
      • At least thousands of relevant customers.
      • At least ten thousand relevant events in the training window (for example, purchases, emails, sessions).
  3. Check with a simple report
    • Recreate the eligible customers segment in Analytics -> Reports and run a simple count:
      • If the count is close to 0, your prediction may have no one to learn from.
      • If the count is extremely high (for example virtually all customers), the model may be working with a very large dataset, which can increase processing time.

2.2 Verify you are using the right template and events

Some prediction templates expect specific events and properties. For example:

  • Purchase prediction – expects purchase-related events (for example orders, transactions).
  • Churn prediction – expects activity/inactivity data over time.
  • Open email prediction / Optimal send time – expects properly mapped email events with a status like “delivered”, “opened”, and so on.

Checks to perform:

  1. Template vs. use case
    • Confirm that you selected the template that matches your use case.
    • Avoid using an email-based template on non-email events, or vice versa.
  2. Event mapping
    • For email-based predictions, confirm that:
      • The correct email campaign events are mapped.
      • The template is using the recommended email status property (not a generic field like action_type).
  3. Is the target behavior clearly defined?
    • Check that the “What do you want to predict?” (target) is properly set and corresponds to a real, well-recorded event.

2.3 Review “Eligible customers” and filters

The Eligible customers filter controls which customers are used to train and score the model.

Things to verify:

  1. Not too narrow
    • Very strict combinations of conditions (for example many AND filters, very short recent time windows, niche segments) can reduce the eligible population to almost zero.
  2. Not unnecessarily huge
    • If everyone in your customer base is eligible and you have a very large database, training may take longer.
    • Consider whether you can focus on a more relevant segment (for example, active customers or customers in a specific market).
  3. Sanity check in a report
    • Apply the same filter in Analytics → Reports to see how many customers match.
    • If it’s 0 or extremely low, adjust the filter.
    • If it’s massive, consider narrowing it slightly.

2.4 Review the prediction time windows

Most predictions have a configuration for:

  • A target window (for example “will purchase within the next 30 days”), and
  • A training/feature window (for example “use behavior in the last 60 days to predict”).

Check that:

  • The windows align with your use case (for example not predicting 365-day behavior with only a few days of data).
  • The feature window is long enough to contain meaningful historical behavior.
  • For “now” or in-session predictions, the feature window appropriately covers recent activity (for example, the last X days including today).

If any of the above appears clearly wrong, adjust the configuration, then restart the prediction (see next step).

Step 3 – Try a clean restart

If your prediction has been in Queued status for a while and the configuration seems reasonable, try restarting it:

  1. Stop the prediction
    • Go to Analytics -> Predictions -> [Your Prediction].
    • On the Results tab, click Stop.
  2. Wait briefly
    • Wait a short time (for example 1–2 minutes) to ensure the stop command is processed.
  3. Start the prediction again
    • Click Start again.
    • Return to the Results tab and monitor the status.

After a restart:

  • It is normal to see Queued again initially.
  • The prediction should again move into processing/training within a reasonable time.
  • If the prediction remains in Queued for more than ~2 hours after a restart, proceed to the next step.

Step 4 – Optionally create a simpler test copy

As an additional diagnostic step, you can create a simplified copy of the prediction:

  1. Create a copy
    • In Analytics → Predictions, open the existing prediction.
    • Use the copy/duplicate option (if available in your UI) to create a new prediction based on the original.
  2. Simplify the setup in the copy
    • Use shorter time windows (for example, shorter lookback period).
    • Narrow the eligible customers to a more focused but still sizable segment (for example active customers only).
    • Remove obviously unnecessary features (for example, certain attributes that aren't critical for the test).
  3. Start the test copy
    • Start the new prediction and monitor the Results tab.

Possible outcomes:

  • If the simplified copy finishes normally, but the original still gets stuck:
    • The original configuration may be overly heavy or complex.
    • You can adjust the original based on what worked in the copy.
  • If both the original and the copy stay in Queued for a long time:
    • This strongly suggests a backend/platform issue, not something you can fix in the UI.
    • You should contact Bloomreach Support (see next section).

Step 5 – When to contact Bloomreach Support

Contact Bloomreach Support if any of the following is true:

  • Your prediction has been in Queued continuously for more than 24 hours and:
    • You have verified the configuration and data (as described above), and
    • You have already tried to Stop and Start the prediction again.
  • A simplified test copy of the prediction also remains in Queued for an extended period.

Information to include in your support request

Providing the information below will help Support investigate and resolve the issue faster:

  1. Prediction URL(s)
    • Copy the full URL(s) from your browser’s address bar for:
      • The original prediction.
      • Any test copies you created.
  2. Project and environment
    • The project name and environment (for example: production/test).
  3. Timeline
    • When you first started the prediction.
    • Whether, and when, you stopped and restarted it.
    • How long has it been in Queued since the last restart.
  4. Prediction details
    • Which template or prediction type you are using (for example purchase prediction, churn, optimal send time, custom prediction).
    • Any notable aspects of the configuration:
      • Training window and target window.
      • Eligible customers filter (for example rough size of the eligible segment).
      • Any recent changes to data mapping or tracking events that might be relevant.
  5. Scope of the issue
    • Whether multiple predictions are affected or just one.
    • Whether you recently experienced other issues (for example delayed jobs, data changes).

Support will then:

  • Check your configuration and data.
  • Verify the status of the underlying training jobs.
  • Involve engineering if needed (for example, to inspect infrastructure or platform-level issues).

Summary

  • “Queued” means the prediction job has been submitted and is waiting to start training.
  • Seeing Queued immediately after starting is normal. Most predictions move into training and complete within 10 minutes to 2 hours (sometimes a few hours for very large setups).
  • If your prediction remains in Queued for more than ~2 hours, you should:
    • Check data availability and configuration (template, events, filters, time windows).
    • Try stopping and restarting the prediction.
    • Optionally create a simplified copy as a test.
  • If the prediction stays in Queued for more than 24 hours, or both the original and a simplified copy remain Queued, contact Bloomreach Support with the prediction URLs and configuration details so we can investigate.

 

Related articles