Commercial Banking and Trade Finance encompasses much documentation.
The ease of customer experience that has evolved thus far in Retail Banking – something such as paperless and contactless UPI transactions – is far away in the case of Commercial Banking.
In commercial banking, customers submit documents for loan applications, account creation, Bank Guarantees or a Letter of Credit. Most of these documents run into 50-60 sheets of paper, details from which have to be manually filled in printed forms or highlighted in templates. In addition, there are supporting documents such as collateral proof, identification documents, and statements that all need to be manually documented and verified.
The banking staff uploads the final set of scanned details. The documents are verified and routed to the following process per the Standard Operating Procedures (SOP). The manually scanned documents are of varying quality with inconsistent imaging.
The entire process takes 4 to 45 working days before the customer request is sanctioned based on the complexity of the use case, or is further delayed on account of some missing information.
A good chunk of manual documentation is involved, so it is prone to errors. It is time-consuming, and there is a severe risk of non-compliance. These would result in operational inefficiencies, regulatory and compliance challenges, dwindling business outcomes and poor customer experience.
The scenario is similar for document processing in other sectors, such as Public Sector Undertakings (PSUs), Real Estate, Insurance, Healthcare, Construction, Manufacturing etc. Consumers may be unaware.
However, today Artificial Intelligence (AI) can be creatively adapted to intelligently process documents in Commercial Banking or Trade Finance, which ensures that manual documentation, which used to take days, can now be processed in minutes. This results in enhanced business outcomes, increased user revenue, and an excellent customer experience.
Using AI, the documents could be automatically picked up from the repository and achieve the desired output in pixel quality and orientation. This will be in image format.
The next step is the extraction of raw text from the image. Next, the text detection is done using Optical Character Recognition (OCR) technique with a contextual understanding and is converted into an identifiable text.
Next is the classification of each document into specific types using Natural Language Processing (NLP). Taking the Corporate Banking experience again would mean classifying the document as a Letter of Credit, Loan application or a Bank Guarantee. The solution can be enhanced with robust, secured transaction technologies like Distributor ledger-based Blockchain.
The AI engine could be trained with algorithms depending on the industry sector and business context, improving its accuracy. To achieve this, the extracted text is run against set templates for each classification to understand the document in its business context, which again uses NLP. This is followed by separating each word into individual tokens, removing no-meaning words using the stop words repository, identifying keywords from the synonyms repository, tagging shareholders and scope from the opportunity repository and understanding relevance from the sentiment repository.
Exceptions would be thrown while comparing the template. Advance rule engines must be set based on exceptions for automated processes. Exceptions must be set up according to the classification type. Keeping with corporate banking, this could be a missing keyword (like “working capital”), an insurance alert, an Incoterm confirmation, an adverse payment term, a date that is outside of the acceptable range, an address alert, a date mismatch, an amount mismatch, an amount that is over the allowable limit, a tenor mismatch, etc.
A new ticket can automatically start the entire procedure. If no exceptions are discovered, it can proceed without intervention and be forwarded to the final authority for sanction. Depending on certain exceptions, some tickets may be automatically refused. Additional tickets with the appropriate exception(s) can be put in the proper buckets so the appropriate authority can immediately act. The system can point out the exception.
Combining AI, ML, NLP and Blockchain and triggering an integrated workflow would enhance business outcomes significantly while providing an exceptional customer experience. It will also ensure increased process compliance. Employee morale would also be boosted as they would be freed from repetitive manual work.
(The author is Mr. Subramaniyan Neelakandan, Founder-Director, Impactsure Technologies, and the views expressed in this article are his own)
