SMART PAIMENTS = AI IN PAYMENTS

AI in Payments = Smart pAIments

Last SIBOS the buzzwords were ISO, Metaverse and Embedded. It was forecast that “Embedded” will be a $7Tr market by 2030 and payments will be 60% of that and on the Metaverse, JPMC predicted it to be a  $1Tr market and HSBC also made announcements of investing in this area.

This time round, my guess is that the trendy topic will be around AI. With generative AI having taken the world by storm, everyone is betting big on AI. @Mckinsey in a 2023 report forecasts generative AI to, yes, “generate” $2.6 trillion to $4.4 trillion annually across some 63 use cases.

Within Payments, there is ample scope for optimization using Machine Learning, Business Intelligence, (conventional) AI and Generative AI. I would divide this into 4 broad areas during payments processing:

a)       Payments Initiation use cases such as using AI to determine the fastest and cheapest payment method for a corporate treasurer, recommending optimum FX Rates for a cross border fund transfer, or recommending best source of liquidity across multiple accounts of a corporate to make a payment (thereby reducing overdraft charges ). At Initiation, AI can also be used for Beneficiary name, account and currency validation.

b)      Payments Execution use cases  to  determine best and least cost payment route across a network of correspondent banks,  enrichment of bank codes from bank name and address to a BIC Code, enrichment of debtor and creditor address from unstructured format to structured formats

c)       Payments Screening for fraud, AML and sanctions based on multiple patterns such as smart name matching, Max transaction amount, Max transactions per file, Max files per day / per hour, Max transactions per day, Large value / count of payments to a new beneficiary, Payment initiation from a new geo-location/IP, Payment initiation by Time of day / day of the week pattern recognition

d)      Payments Monitoring – here AI can be used for monitoring logs across various workflows and systems, pile-up of transactions in a particular queue, longer than expected response time from downstream systems and triggering alerts and notifications and possibly even taking corrective action ( self-healing based on similar situations in the past )

Then there are two other areas which are outside the above “payments execution cycle” where AI can be used

a)       Payments Analytics where AI can be used to analyse historic data and come up with trends. For e.g. a RM could cross sell a receivables finance product where a trend indicates a large number of incoming payments with Paid-Date > invoice-due-Date. AI can find patterns very useful for different stakeholders such as the bank relationship manager, bank operations manager, product manager or even for the corporate

b)       Payments Support Chatbot – where generative AI tools can be used to provide guidance and support to the end user. These tools can be used by corporate treasurers to find out patterns in payor behaviour and then offer Virtual Accounts or Request to Pay to its payors to prompt early payment. These tools can also be used to carry out payment fee analysis and settlement time analysis so that appropriate banks / payment methods can be used.

Banks planning to leapfrog paiments ( AI in Payments) must decide what are the most important use cases for them ( internal stakeholders – operations, product, relationship managers   )  and most important use cases for their clients ( corporate treasurers, AP, AR functions etc ) and then detail out those use cases. Once that is done, it is important to have a large data set to train and fine tune the AI model so that the results are (near) consistent. Once the model has been trained, a good amount of testing is also needed. Do not get carried away by the hype. Remember Generative AI tools as of date are still in evolution phase, are subject to hallucination and hard to “test” but still have a very meaningful role to play. The key is to be clear whether you want to use conventional AI and ML or generative AI depending on the use-case.

Clearly this space is evolving at a very rapid pace and early adopters will stand to gain.