Automation is a hot topic. When done properly automation allows a more efficient work flow enabling more work done at a lower cost While this is bad news for labor, its great news for businesses and their investors. In this article, I intend to discuss the automation of signature verification for banks.

History

Before computers, banks maintained cardboard cards for each account that had the signatures of the account holders. While this was practical back in the day, as customer numbers increased, and efficiency and security was paramount, banking software incorporated features to help signature verification around 1987.

When entering transaction details, using the account number, the account holders signatures will be presented to the teller from a central database. A quick visual inspection is done before finalising the transaction. Large transactions would be scrutinised further by a supervisor.

People have been hoping for a day computers could automate this process so the teller and his supervisor can have some peace of mind.

Software Tools For Signature Validation

A signature is unique to a person. Yet a given person will never make 2 identical signatures as there will always be variations and thus a simple image comparison cannot be used to verify the signature. To verify with a high confidence level, various approaches have been developed to extract and evaluate both the static and dynamic (biometric) characteristics of handwritten signatures.

Dynamic being the scenario where the user is live inputting the signature (ex: using a tablet), banks would often use static approach as they will be verifying signatures that already have been made (ex: on paper - credit/debit slips) which purely involves image processing.

These methodologies can be used to quantify the confidence of the accuracy of the given signature. Machine learning techniques will require a set of training signatures, at minimum 10 and more if in order to increase accuracy, and be done in a per customer basis. Given the account details the models can be loaded up and checked against.

For a good set of training data, these AI modules could reach almost near 100% accuracy.

The Concerns

The concept of automation is to increase efficiency and the overall quality / security. While software could, in the right circumstances, achieve accuracies even the best of human forensics would find difficult to achieve, its practical applications I believe are questionable.

People who do online banking will not require a signature. The use of the username/password coupled with a one-time-password does verify a persons authority over an account. Globally the trend is that people are moving from traditional banking to digital forms of banking.

People who do traditional banking would use paper based forms and cheques. A manual image comparison by the teller using an image obtained from a central database would only take a couple seconds. Should a digital form of signature verification (AI and machine learning) is to be used, the signature will have to be scanned. This adds to the workload rather than reduce it. It also adds an additional point of possible failure, reduced efficiency and most definitely additional cost.

Conclusion

As the society embrace digital banking signatures will become less prominent in banking applications. I have my self gone from writing an average of 5 cheques a month to none for over a year. All of my transactions are done online. They are faster, convenient and more secure.

Enforcing a AI - machine learning based signature verification will involve challenges. At one hand we will need a set of signatures from the client to train the modal. This can prove a challenge as some customers even struggle to place a single signature. In the other hand to verify a signature in a paper using software will require scanning the document. This will slow down the general flow.

While technically a better accuracy can be achieved using a digital signature verification process, in my opinion alternative approaches should be used. For example we could force a complete digital transition by introducing more robust smart phone based banking solutions similar to WeChat in China.

This is my personal opinion on the matter based on my experience and is an ongoing investigation. I intend to reach out to tellers, managers and IT professionals working in the banking sector to investigate the current use of signature verification systems and their opinion on the topic.