Skip to main content
Skip table of contents

Understanding Confidence Scores in Express

With a document open in Express, hover over any field that is pre-populated. If the field was predicted by AI, the application will display an AI confidence score. This score, in a tooltip text box, shows how reliable the field’s information is, offering you insight on whether to approve or to change the data. 

image-20250811-170619.png

Hover over a field to reveal the confidence score.

Along with the score, fields that don’t meet the configured confidence threshold settings are highlighted in yellow, as warnings.

image-20250811-172729.png

If the threshold is not met, a warning displays

After you approve the document, Conexiom’s AI learns from the results and applies that knowledge to future documents.

Using Confidence Scores

The AI’s prediction of an entry in the field can, at times, be inaccurate, so the confidence score gives you a trustworthy measurement of how well the AI performed. It is the result of calculating how often predictions were correct in similar situations, based on real-world data. In this way, it can translate into how confident you can be in the data.

A score of 90% is the default threshold setting for mandatory fields and equates to a high level of confidence. A score of 60% is the default threshold setting for non-mandatory fields.

Setting Your Confidence Score Thresholds

You can set confidence score thresholds to customize document automation. If a prediction falls below the threshold, Conexiom can alert you and stop the automation, helping you avoid errors. To learn more about setting confidence score thresholds in Conexiom, visit Setting up Autopilot, AI Confidence, and AI Prediction Features

Identifying AI-predicted Fields

To see if a field is AI-predicted, hover over the field name. If it’s an AI-backed field, a tooltip displays. You can also find the full list of AI-predicted fields here for purchase orders (PO) and here for invoices (INV).

JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.