---
sourceDocument: Brazil Enable AI
sourceDocumentLink: https://www.servicenow.com/docs/r/intelligent-experiences

 Release :

    - brazil

ft:locale :

    - en-US

ft:publication_title :

    - Brazil Enable AI

ft:clusterId :

    - platai

bundleId :

    - platai

workflow :

    - Platform


---

# Test time explainability

# Test time explainability {#ariaid-title1}

Release version: Brazil  
Updated September 10, 2026  
![](https://www.servicenow.com/docs/portal-asset/ico-clock) 2 minutes to read  
Understand the reasons your machine learning solution made a specific
prediction with Test time explainability, which returns a list of input features that influenced the prediction.

## Overview of Test time explainability {#predictive-intel-test-time-explain__section_sfm_swm_hkc}

When Test time explainability (TTE) is enabled on a classification solution, every prediction is accompanied by an explanation that lists each input feature along with a contribution score. Features with the largest contribution
scores are the strongest drivers of that individual prediction.

## Benefits of Test time explainability {#predictive-intel-test-time-explain__section_gy5_31l_jkc}

* Transparency: Explanations turn scores into confident decisions.
* Accountability: Each prediction comes with a per-decision record of the factors that combined to shape the result.
* Faster troubleshooting: Unexpected predictions are easier to investigate using the list of factors and their influence.
{#predictive-intel-test-time-explain__ul_xgq_l1l_jkc}

## Considerations for using Test time explainability {#predictive-intel-test-time-explain__section_nrp_lgl_jkc}

Be aware of the following considerations when using Test time explainability (TTE).

* TTE supports only workflow Classification solutions.
* TTE supports only the xgboost classifier method. Don't change the classifier_method when the TTE flag is enabled.
* TTE is turned off by default. You must set TTE to true during training and predicting.
* TTE can be applied to nominal (categorical) and numeric columns only, not text columns. If input features consist of a mix of text, nominal, and numeric columns, the text columns are dropped. If the input features consist of text columns only, the solution training fails.
{#predictive-intel-test-time-explain__ul_hy5_lhl_jkc}

## Invoking Test time explainability {#predictive-intel-test-time-explain__section_xzq_twm_hkc}

Test time explainability is set in two API calls: at the training stage, and at the prediction stage. When TTE is enabled, the prediction response includes an `explanation` object.  
The following code snippet shows an example of defining a classification solution for training. The parameter test-time-explainability is set to true (default is false).

    var mySolution = new sn_ml.ClassificationSolution({
        'label': 'telecom_customer_churn_with_tte',
        'dataset': myData,
        'predictedFieldName': 'churn',
        'inputFieldNames': [
            'age',
            'contract',
            'monthly_charge',
            'tenure_in_months',
            'online_security',
            'online_backup'
            // ... additional input fields
        ],
        'test-time-explainability': true
    });

The following code snippet shows an example of requesting explanations at prediction time, after the model has trained successfully. The option test_time_explainability is set to true (default is false).

    var input = new GlideRecord('u_telecom_customer_churn_train');
    input.get('07601c45ff9232107334ffffffffff38'); // record must not be empty

    var options = {};
    options.apply_threshold = "false";
    options.test_time_explainability = "true";
    options.top_n = 100;

    var solutionName = sn_ml.ClassificationSolutionStore.get(
        "ml_sn_global_global_telecom_customer_churn_with_tte"
    );
    var solutionVersion = solutionName.getLatestVersion();
    gs.info(solutionVersion.predict(input, options));

The following snippet shows an example of the explanation in JSON format (trimmed for readability).

    {
      "07601c45ff9232107334ffffffffff38": [
        {
          "confidence": 99.53,
          "threshold": 1.61,
          "predictedValue": "0",
          "explanation": {
            "prediction": 0.0047,
            "base_value": -0.1806,
            "all_contributors": [
              { "feature": "contract",          "raw_value": "Two Year", "encoded_value": 2.0,  "shap_value": -3.4014 },
              { "feature": "monthly_charge",     "raw_value": "58.55",    "encoded_value": 0.212, "shap_value": -0.6699 },
              { "feature": "tenure_in_months",   "raw_value": "29",       "encoded_value": 28.0,  "shap_value": -0.5553 },
              { "feature": "age",                "raw_value": "68",       "encoded_value": 49.0,  "shap_value":  0.3234 }
            ]
          }
        }
      ]
    }

Following are the definitions of the fields in `explanation`.{#predictive-intel-test-time-explain__table_aqk_dpl_jkc__entry__2}

| Field | Definition |
|-|-|
| prediction | The model's output for this record after all feature contributions are applied. |
| base_value | The model's expected result before any feature contributions. |
| all_contributors | The full list of input features and their contribution to this prediction. * feature: The name of the input field. * raw_value: The original value from the record (for example, "Two Year" or "58.55"). * encoded_value: The internal numeric representation of the original value used by the model. * shap_value: The contribution score. A positive value pushes the prediction higher; a negative value pushes it lower. The larger the magnitude, the stronger the influence. {#predictive-intel-test-time-explain__ul_imx_vrl_jkc} |
[Table 1. Fields in the explanation response]

{#predictive-intel-test-time-explain__table_aqk_dpl_jkc}The contributors are ordered by influence, so the features at the top of `all_contributors` are the strongest drivers of the prediction. Starting from the `base_value`, each `shap_value` influences the result up (positive value) or down (negative value).
**Related tasks**   

* [Configure XGBoost for classification or regression solutions](https://www.servicenow.com/docs/8J9l0_Xr2oWxQtLOP4b4fw "Apply XGBoost encoding to optimize the training for your classification or regression solutions.")  
**Related topics**   

* [ClassificationSolution - Global](https://www.servicenow.com/docs/access?context=ClassificationSolutionAPI&version=brazil&pubname=brazil-api-reference&ft:locale=en-US)