Insights

The Insights page gives you a full performance analysis of a Trustfull model against your labelled data. It shows how well the model's scores matched real-world outcomes across your traffic, and lets you simulate the effect of changing the reject threshold on fraud detection, approval reduction, and false positive rates.

📘

Generating insights requires at least one Training Set uploaded under the same App Key. See Training Sets for instructions on how to upload one.

How to Run Insights

Generating an insight takes three steps: select the app key and training set you want to analyse, choose the model to run the analysis against, and confirm. Results are ready in the Insights list once processing completes.

Open the Run Insight modal

From the Insights page, click Run Insights in the top right corner.

⚠️

Generating insights may take up to 5 minutes. Once completed, the results will appear in the Insights list.

Select App Key, Training Set, and Model

In the Run Insight modal, provide:

• Select App Key: the app key associated with your training set
• Select Training Set: choose from the training sets uploaded under that app key
• Select Model: choose the Trustfull model to run the analysis against

Only models matching the product of the selected training set are available. For example, if your training set contains Session lookups, only Session models will appear.

If no training sets are available for the selected app key, you will be prompted to upload one first.

Run and view results

Click Run Insight. Once processing is complete, the insight appears in the list. Click the magnifying glass icon to open the full insights page.

Reading the Insights Page

At the top, a summary bar shows the training set name, total records, number of approved and rejected records, upload date, and the model used.

Reject Threshold and Key Metrics

The Reject Threshold slider controls the score below which Trustfull recommends taking action (rejecting or requesting additional verification). The standard threshold is 450, meaning any lookup with a score below 450 is flagged for review.

The three metric cards update as you move the slider:

• Fraud Detection Rate: the percentage of fraudsters detected out of all fraudsters in the training set, for the selected threshold
• Approval Reduction Rate: the percentage of total traffic blocked for the selected threshold
• False Positive Rate: the percentage of legitimate records flagged as fraudulent out of all OK records for the selected threshold

To simulate a different threshold, change the Scorecard setting from Standard to Custom and drag the slider. For example, raising the threshold from 450 to 550 will typically increase the fraud detection rate, but will also increase the approval reduction rate and the false positive rate.

Other Charts

The insights page includes several charts to help you interpret the results

• Score Distribution: three charts (Fraud Detection Rate, Approval Reduction Rate, False Positive Rate Analysis) plot each metric across the full score range (0 to 1000), with a vertical line marking the current threshold
• Sample Distribution Across Score: a scatter plot showing how OK and KO records were distributed across the score range. KO records should cluster on the lower end, OK records on the higher end. Click any dot to open the corresponding lookup.
• Score Cluster Distribution: a breakdown of how records are distributed across score clusters (Poor, Bad, Moderate, Good, High)
• Scoring Model Accuracy: a matrix showing, for each cluster, the percentage of OK and KO records. For example, if 96% of records in the Good cluster were labelled OK, the model is correctly identifying low-risk users in that range.
• False Positives Analysis: a table listing records flagged as fraudulent but labelled OK, with customer ID, reason codes, and score. Use this to identify patterns and refine your threshold or model configuration.
• Feature Importance and Feature Distribution: the signals most associated with rejected applications ranked by impact, and how those features are distributed across approved and rejected records.

Managing Insights

From the Insights list, use the icons on each row to:

• View (magnifying glass icon): open the full insights page for that entry
• Delete (bin icon): permanently remove the insight

📘

You can filter the list by Training Set or Model using the dropdowns above the table.