A recent FT article – ‘AI in banking, the reality behind the hype’ – discussed the potential of AI to revolutionise banking. But beyond the use of chatbots to automate customer interactions and machine-learning to monitor customer behaviour patterns in the fight against fraud, is there an opportunity to deploy AI in payments assurance testing ?

The cost and inefficiencies associated with manual testing are acknowledged in the Industry and will only continue to intensify unless a different approach to the issue is adopted – eg the deployment of innovative new technologies. Analyst IDC estimated that financial institutions will spend more than $12 billion by 2019 on transforming their payment systems, and a Capegemini report noted that 31% of IT budgets are spent on testing – suggesting a $3.7 billion testing cost-save opportunity the likes of AI in payments alone.

The successful use of AI across a broad range of industries, including healthcare, construction, mining and agriculture, demonstrates both its capabilities and advantages. For example, AI can be used to:

• undertake complex problem solving through the analysis of massive amounts of data
• remove human-error from processes and their oversight making things safer and more reliable
• automate a range of processes thereby reducing cost and increasing efficiency.

All these advantages would clearly be of benefit to payments assurance testing. So how might an AI-based approach to testing look ?

AI would be at the heart of a completely re-engineered approach to both systems development and testing. As developers wrote the code for new products, AI systems would be running in the background automatically producing Use Cases by which the new code could be tested. The Use Cases would build-in the appropriate standards and policies of the institution ensuring a consistent and compliant approach.

As to the testing itself, this would be based on AI-generated test code to test against the specific acceptance criteria of the Use Cases concerned. Then, once testing was underway, machine learning would be used to automatically interpret the testing results - removing the need for human testers. The machine could be taught to rank the severity of the errors and issues it discovered in the testing for reporting purposes and even suggest the specific modules of the code that caused them.

Further, the ability of AI to cope with vast amounts of data, enables more complex end-to-end testing and ‘what if’ scenario modelling to be undertaken automatically. Thus allowing testing to be ‘super-charged’. Automated user acceptance testing would be able to survey a much greater range of business situations moving from the basic testing of individual elements of a new system to AI led quality assurance on a full end-to-end basis.

The complex nature of end-to-end payments testing involving so many bank systems - customer initiation via remote devices/apps, Open APIs, two-factor authentication of client identity/authority, credit risk systems, interfaces to accounting systems, payment routing and transaction formatting, gateways to payment infra-structures, intra-day liquidity management, reconciliation, query handling etc – determine that institutions should take advantage of the application of the most up-to-date testing capabilities – including, in particular, AI. Otherwise, the costs of end-to-end testing of payment changes will continue to escalate and it will remain a difficult business to provide the assurance required by the institution’s executive management and their regulators.

Whilst such a use of AI is the future, building an AI response to the current disconnected infrastructure would be a massive task for most companies. To secure the ability to benefit from what the future will clearly deliver, banks need to make necessary investments in the way they go about delivering assuredness now, and avoid building more and more silos with no clear strategy or direction.