back  Back to Case Studies

12 June, 2022

At one of the largest Indian bank, we process lakhs of cheques at the rate of 20 cheques per second.

For any bank, Cheque clearing is a very important and highly resource intensive process. On any normal day, a leading bank can receive anywhere between 3-5 lakhs of cheques for clearance. Being an intricate process, it involves multiple levels of screening and vigilance to ensure regulatory guidelines are met. For all these reasons, major parts of this process are manual in nature till date. Guidelines involve verifying account information and some technical details present in the cheques. Some of the technical verification guidelines are as below:

  1. The Amount in words needs to be same as Amount in figures
  2. Cheque date must be within past 90 days from the current (presenting) date
  3. Signature in cheque must match, with a high degree of confidence, to the signature present in core banking database of the corresponding account

Use case

The primary purpose of an AI based cheque clearance system is to reduce the processing time while also making the process less prone to errors and possible frauds due to collusion.

Arya’s module works on automating the process with the use of six different AI models for the various fields including Date, MICR, Amount in words and figures, Account Number and Signature extraction and verification. Each AI model is trained for high precision results using historical data of cheque images as well as signature snippets as provided from the database for signature matching.

  1. Extract all the handwritten information on the cheque
  2. Extract the signature present in the cheques
  3. Validate the information against the business rules deemed by the organisation
  4. Verify the signature present in cheque with that of in the Organisation's database

Current Process

  1. Cheques in batches are uploaded at CHI servers of clearing banks via SFTP from NPCI servers
  2. All clearing banks have the CTS system where the cheques with its details are broadcasted/published for employees/tellers to verify at different levels for clearance
  3. Most of the process of validating cheque details is being done by human resources
  4. A manual review of high volume of cheques is not only time consuming, and since most verification happens at odd hours of the night time, it is also prone to human errors

Impact

Arya’s module has increased the efficiency of processing huge volumes of cheques. Each model generates a confidence score for it’s prediction of the field value. Based on these confidence scores, and after applying various business level validations, a sizable volume of cheques are cleared with minimal human intervention. Moreover, a highly precise signature verification model also helps in reducing the number of frauds.

  • Processing time of 15-20 Cheques per second on a standard production infrastructure
  • Processing cost to be reduced by upto 55%