AI helps treasurers keep pace with changing demands of role

By understanding why a client chooses certain services, banks can make better recommendations, acting more as an adviser and less about pure execution, she explains.

AI can help banks and their clients in a myriad ways, such as reducing operational costs, ensuring regulatory compliance or improving analytics, both to better understand existing data and produce better forecasts.

Leap in the dark

However, there is a leap in the dark required. The technology is new and in some circumstances unproven. The return on investment is hard to measure and complex, being part cost-growth minimization, part indirect revenue and part lower fraud losses, says Grealish.

Given the high rates of CFO turnover, this can make it particularly difficult to plan strategically and make the right long-term investments.

However hard it may be to measure with precision, the case for investing in AI seems clear.

Dean Henry, head of innovation in global transaction services at Bank of America Merrill Lynch (BAML), says: “We look for three things in any kind of technology, not just AI. We want it to make client interaction easier, to increase efficiency and improve risk management.”

Without coming up with a single number to quantify it, AI can clearly deliver these benefits in a number of functions. Cash forecasting, for example, is a traditionally manual process, involving examining paper statements for different account details, as well as aggregating things such as purchasing orders. AI can do this better than any human, making it considerably more efficient.

Henry says: “AI has an incredible ability to synthesize all this information and make insightful predictions about future cash flow.”

However, the progress banks have made in developing and using the technology is unsurprisingly inversely correlated with the technical complexities involved.

Synechron’s Gupta says: “At the simpler end of the scale, AI can help in low-value, high-volume areas [such as] data entry, automated decision-making for reconciliations and identifying trading breaks. At the medium to complex end of the spectrum are regulatory reporting and predictive analysis capabilities.

“The reporting formats are standardized, but data calculations and manipulations underlying the reports are not. Ultimately, self-learning machines will be able to help with that.”

Cautious roll-out

Celent’s Grealish says banks have made less progress developing customer-facing AI applications.

Having studied a number of banks’ AI investments, she found that “most customer-facing applications are just heading to the launch pad, which is not surprising given the complexities involved in scaling on the commercial side compared to the retail side.”

BAML’s Henry agrees, saying: “We are more inclined to take risk when we can control the impact, which is easier with internal processes than with client facing functions. But we also want to lead and offer our clients services that help them.

“So we are cautious about what we put in front of our clients, we do not offer them anything that has any operational risk attached to it – it can only be about improving efficiency.”

As the bank’s confidence in its AI products increases, BAML will be able to offer them more, he adds.

Pankaj Gupta, Synechron

Perhaps the most obvious use-case for AI in transaction services, at the simpler end of Gupta’s spectrum, is automation of reconciliations. 

Banks routinely deal with a variety of customer errors when initiating payments, and payment repairs and investigations typically account for 75% to 80% of the labour-intensive part of a payment-processing operation, says Grealish.

BAML’s global receivables team launched Intelligent Receivables in August 2017, with a view to enabling straight-through reconciliation (STR) for automated clearing house (ACH) payments for which the remittance information is either missing or received separately from the payment.

More use of ACH and cards has been increasing the processing challenge, with a growing number of clients turning away from banks to fintechs, while regulation is also pushing customers towards greater accounts receivable efficiency.

However, machine learning helped the bank reduce the processing burden by eking out the additional percentage points that traditional technologies could not, pushing STR rates from 10% to above 90%, with direct posting to SAP and Oracle enterprise-resource-planning systems.

BAML plans on enhancing Intelligent Receivables with foreign-currency processing and incorporating it into its virtual-accounts service, notes Grealish.

“Already, its clients are realizing numerous benefits, including improved cash forecasting and the ability to manage trade credit more efficiently,” she adds.

Future of payments

Grealish believes AI will become more important in payments processing during the next five years as the share of machine-to-machine payments and real-time payments steadily rises. And other banks have already developed, or are developing, similar applications.

Panama-based Global Bank partnered with tech vendor Pelican to create PelicanPayments; a system that applies machine learning and natural-language processing to deliver repair logic, customized routing logic and intelligent reporting.

“It learns from the errors operators correct, identifying patterns of repetitive behaviour and then developing auto repair tools,” says Grealish.

Global Bank wanted its AI to auto-repair payment initiation files to push straight-through processing (STP) rates above those of their competitors. Having achieved this, it was also happy to observe AI had increased margins and improved their customers’ experience.

At the more sophisticated end of Gupta’s spectrum, however, Synechron’s RegTech Accelerator program can help address variances in capital calculations, using machine learning (ML) and natural-language generation (NLG) for computation and commentary generation.

“Algorithms also improve with time through creation of continuous improvement loops,” says Gupta.

Citibank’s engagement with AI has also attempted to improve the insights it can deliver to clients.

Grealish says: “Citi has produced some valuable AI applications, including payment-outlier detection, cash optimization and forecasting, working capital optimization, and contextual insights from sales and payments data. It is working on applying natural-language processing to real-time contract negotiation and building a contextual recommendations engine.”

For example, a client seeking to expand into a new country might find Citi’s insights helped inform its decision, giving its access to unique analysis based on proprietary data, such as trends in payroll volumes, says Grealish.

“Citi is well-positioned to be a leader in data analytics that is business and client-driven and differentiate itself based on superior customer and employee experience,” she adds.

Dean Henry, BAML

Of course, in providing advice ahead of a potential move into a new market, banks are not breaking new ground, as much as improving the advice they have always given. Even the AI component is not revolutionary, says Henry. 

“It is a common misunderstanding of AI that it is all completely new,” he says. “Think about AI as four building blocks: robotic process automation; analytics; cognitive; and conversational.

“The first – robotic process automation – has been around for years, and we are doing a lot with the second as well. Intelligent Receivables uses the first three. Our retail bank is leading our experimentation in the fourth area with Erica [the bank’s virtual voice assistant], but this technology is nascent.”

Gupta adds: “AI capabilities are showing progressive improvement. Similar to humans getting better at their jobs with experience, so will machines enabled by richer data sets and feedback loops.

“Banks have started with simpler internal applications of AI; however, they are progressing to more complex and outward facing functions, with technology enhancements and sentiment improvement.”

He adds: “This is also coupled with significant improvements in convergence of technologies such as AI, ML, optical character recognition, natural-language processing and NLG in disrupting entire value chains.”

Mistake

However, it is a mistake to expect AI to completely transform the banking business, says Henry.

“Transaction banking is highly regulated,” he says. “It is evolving, but it can only do that within the confines of rules governing flows of money.

“What is really transformational is not so much banking itself but our clients – the end-consumers – and how they use technology, of which AI is just one part. People are more engaged with the digital world now, our clients’ businesses are also increasingly digital. So all banks are doing is trying to find ways to deploy the regulated, core services they have always provided, within that digital environment.”

That includes a human component, which is not likely to be driven out of banking any time soon, says Gupta.

“While technology will provide the base processes in these cases, we should expect a layer of human intervention to provide failsafe mechanisms in the event of technological failures,” he says. “All banks need to actively develop not just base processes, but also strong control processes for such occurrences.”