Document Processing Solutions and Significance of STP

In a typical data capture process, there is a maker-checker model adopted to ensure that the data accuracy is maintained at ~100%. The first actor is the maker who reads data from a document and captures (enters) into a data capture form or application. Next, this captured data is subjected to validation by a checker who also corrects if there are any anomalies. In certain cases, based on the data accuracy need, businesses may choose to have more than one checker process.

An experienced maker takes approximately 0.8 seconds per character for data entry. Examples: Simple data fields such as numerical cheque amount, or first name / last name may take 3-5 seconds. Lengthy text fields will need more time - examples: address, drug dosage, line items in invoices, cheque amount in words.

For accurately captured data, the checker agent’s effort is limited to visual validation. When data is anomalous, the checker has to spend additional time to correct it. This makes the checker process heavier. Often the checkers find it easier to just erase the data and enter it. This is because making a correction involves more effort and possibly more keystrokes than a key-in.

The First Time Right (FTR) principle plays a significant role to determine the overall efficiency of the process. If the maker does good during the data capture, the checker’s effort is limited to validation only, thereby, enhancing the collective productivity of the team. When an automated solution is employed as the maker, higher accuracy of the data fields translates to lesser validation effort by the checker.

Automated data capture solutions have brought many interesting implications unto the table. Unlike human agents, robots (software solutions) have the capability to capture data and indicate a confidence associated with such capture. When this confidence score is very high, it can be assumed that the machine has done a perfect job and eliminate the checker step for such data fields. Since such a data is ready to be used without any human touch during the capture process, it’s referred to as a straight through processing (STP). As the STP rate of automated capture increases, the cost of data acquisition falls.

Below screenshot of a typical data capture application shows the captured data fields with a color-coded confidence configured basis the accuracy.

• Green indicates high confidence and can be configured for remaining in or out of human validation on the data panel thus enabling STP
• Amber indicates medium confidence when the accuracy threshold is lesser than high-confidence and the data may or may not be accurate thus calling for a human correction
• Red indicates least confidence and mandatorily needs human intervention

Businesses should be wary of being misled by claims of high accuracy on character-level recognition. Smart solutions should be able to recognize characters in a data field with high accuracy, be able to demonstrate business confidence with an STP flag for each captured data. Additionally, the solution should be able to learn from a small set of seed samples and enhance its capability on the job; the solution should be able to receive feedback from human checkers on mistakes committed and enhance themselves. Such features are no more a wish list but are fast becoming realities.

In summary, a smart automation solution dealing with capture of data from documents should provide the below features:

1. Recognize all characters of a data field correctly with confidence, thus eliminating it from manual inspection
2. Provide a flag to indicate the fields that should be inspected by humans to ensure accuracy
3. Learn from mistakes and improve on the job