Automated Handwritten Document Data Extraction

large volumes of forms in multiple formats with different handwritten texts, across their back-office processes. Going through each form, identifying different styles of handwriting manually, reading checkboxes and recording the data was tedious and error prone

~10,000 handwritten forms were being processed manually each month. The entire process was labour-intensive and involved eyeballing of forms, forcing employees to intervene rather than spend their time better by focusing on other crucial tasks.

Processing a large quantity of handwritten 401(k) forms manually with human intervention at multiple points results in longer processing times, higher error rates and delayed payments – leading to unhappy customers.

The handwritten forms were processed and maintained, which was proving to be expensive and increasing complexities in extracting data accurately from formats such as checkboxes.  

A private cloud deployment of the Botminds AI platform was done to keep documents secure in the client’s environment.

The scope of the initial automation was identified - two forms that contributed to over 55% of volumes and the data from 40-45 fields were extracted for downstream processing

50-70 samples were selected for training the model and this was sufficient to generate predictions with over 80% accuracy.