Extract accurate data from Tables

Problem

In 2020, Eigen needed to extract data from tables in documents, which was challenging as their initial machine learning models weren't trained to process tabular data effectively.

Solution

Led the user experience for a new table extraction workflow, introducing:

• Multiple extraction modes:

  • Table classification

  • Mixed classification

  • Table point extraction

  • Mixed point extraction

• Support for full-table and cell-specific extraction

• Confidence scoring to indicate extraction reliability

Designed interfaces to:

• Configure extraction type

• Preview confidence levels

• Evaluate outputs with precision/recall metrics

Users reported that the new workflow felt:

• “Clearer and more trustworthy”

• “Much easier to decide whether to accept or edit extractions”

Feedback highlighted how confidence scoring helped reduce anxiety around ML reliability.

Outcome

By reducing manual review, the feature:

• Supported efficiency gains for customers processing high volumes

• Contributed to higher retention and satisfaction on key enterprise accounts