Matching incoming payments with invoices has long been a frustration for companies with many valuable hours being spent trying to determine who’s paying for what. However, artificial intelligence (AI) and machine learning solutions are starting to emerge that claim they can combat these treasury headaches.
“Incomplete remittance information typically leads to an arduous and costly reconciliation process”, argues Rodney Gardner, head of global receivables in global transaction services at Bank of America Merrill Lynch (BoAML).
Data analytics AI technology is still in the early stages of deployment, but it is already set to transform financial markets and could act as a key differentiator in performance, Matthew Hodgson, CEO of Mosaic Smart Data tells GTNews.
With take-up of AI on the rise, PricewaterhouseCoopers (PwC) recently predicted that global gross domestic product (GDP) will be 14% higher in 2030 because of artificial intelligence. This is the equivalent of an additional $15.7trn, or more than the current output of China and India combined, according to PwC research.
“Remittance data comes in all shapes and sizes: e-mail; e-mail attachments, portals, back of the envelopes, carrier pigeon”
The real advances for AI in recent years center around the automatic reading of documents, argues Bertrand Cocagne, head of product at Linedata Lending & Leasing. “I’m referring to …systems being able to recognise the structure of documents. For example, advances are being made where AI can recognise the structure of say, a balance sheet document, and automatically exploit the content,” he says.
The most difficult problems in the marketplace today is re-association of the transaction – the credit is in the client’s demand deposit accounts while the remittance data resides elsewhere, argues Gardner. “Remittance data comes in all shapes and sizes: e-mail; e-mail attachments, portals, back of the envelopes, carrier pigeon,” he laughs.
One AI solution that attempts to tackle this issue is BoAML’s AI payments solution, ‘Intelligent Receivables’, currently available in Canada and the US. It uses artificial intelligence and other software to help companies vastly improve the straight-through reconciliation (STR) of incoming payments to help companies post their receivables faster.
Straight-through processing (STP) enables the entire trade process for capital market and payment transactions to be conducted electronically without the need for re-keying or manual intervention. The tool combines AI, machine learning and optical character recognition (OCR) to help business to do accounts receivable reconciliation and payment matching more efficiently.
The proliferation of payment types
One of the biggest challenges when addressing the issues of transaction re-association is the proliferation of payment types, according to Gardner. “There is a press release about a new way to make a payment nearly every week. That’s incredibly exciting but it means the accounts receivables manager sitting in that shared service center better be reading the press and accounting for how that new payment type is going to be processed. If it comes in and falls to the floor, guess what? You’ve got to hire four or more people in your shared service center to manage the new payments. Nobody is talking about that,” he says.
“I am of the opinion that low-value transactions are the best value on the planet bar none. Since low-value transactions are settled the same day, there is no reason why a corporate shouldn’t be telling their payers, ‘here’s how I want to be paid’,” continues Gardner.
Predicting the problems
Reconciling payment deductions is also a headache when matching payments with invoices, for example if a broker takes commission from a deal. Cross-border payments also present additional reconciliation issues, for example if a transaction was billed in dollars but paid in Thai baht. Gardner argues that AI and machine learning can be used to tackle both of these issues.
This is where predictive analytics plays a crucial part. Matthew Hodgson, CEO of Mosaic Smart Data, says: “In many cases, banks are underutilising the data which they generate internally and spending significant amounts on buying market data from trading venues and exchanges.
“However, the internal data is where some of the insights with real competitive advantages are held. Market data is available to anyone with deep enough pockets, but no other institutions have access to this internal data,” he says.
“When I talk to treasurers, they sometimes don’t realise the amount of horsepower that they currently have against reconciling the books”
Using machine learning combined with internal data to reconcile payments can solve a myriad of headaches. For example, if one client, for some reason, continuously forgets to include part of an invoice number, the first time the business can spot the error and correct it using the client’s open accounts receivable file. The next month, if the client does the same bad behavior the machines will spot the pattern and match the correction.
Once predictive analytics is involved, the machine can not only correct the invoice number but also remember and predict when this bad client will pay, for example on the fourth of the month. This information can be fed into the treasury workstation which can then predict when the business will receive the payment which can help treasurers with their cash forecasting.
BoAML will be able to produce these types of predictive solutions in the next couple of years, Gardner expects. “When I talk to treasurers, they sometimes don’t realise the amount of horsepower that they currently have against reconciling the books. They are often in the dark about their accounts receivables from a cash flow forecasting perspective. They have no analytics around predicting their receivables so we’re creating a dashboard for them to help them better manage their cash flow,” he says.
What makes a successful payment product?
When launching new payment technology there are two guiding principles for success, according to Gardner. The first is global consistency.
Secondly, it cannot require the payer to have to do anything different. “From a receivables perspective, the minute you try to change payer behavior, I believe you’ve put everyone at a disadvantage. There are some great tools and solutions out in the marketplace but most of them require the payer to do something different, and you don’t get the up lift,” he argues.
For this reason, he believes bringing AI reconciliation within European virtual account management would be a logical next step for AI corporate payment and invoicing technology. “SEPA was great on the electronic side but not on a straight through reconciliation basis. And why is that? A virtual account works very well – you probably know who paid and why they paid you but the cash application remains the tricky part,” says Gardner.
He proposes marrying virtual accounts with an AI solution. The bot would go into the virtual sub-account, look at the transaction and re-associate it with the remittance data, regardless of where it came from. “It will re-associate those two data points, create a posting file, and boom! You have straight through reconciliation, even in a virtual account scenario,” says Gardner.
A question of trust
One of the key factors holding back AI and machine learning is consumer and user trust.
Cocagne argues that trust can only really come from experimentation and actually seeing the results of successful implementation. “The algorithms used within current AI systems don’t allow for much transparency with regards to how the system rationalizes its decisions, which makes it difficult for the industry to trust the outcomes. The key to building this confidence is really the successful real-life application of AI. With real results and success stories more trust will come,” he says.
Gardner argues this problem is generational. “The older generation sees banks as trusted transactional, information protecting, partners. But I think millennials don’t give it a second thought to adopt new payment innovations. So, it depends on your audience,” he argues. Due to the older generation’s unwavering faith in traditional banks, banks are still key players in AI technology innovation.
Regulators are raising the drawbridge
Regulators are increasingly focusing in on AI payments technology. “Some of the first companies that came into the payment space were out ahead of regulations, and they weren’t necessarily regulated as banks,” Gardner tells GTNews. “Now there’s a catch up by the regulators so the barriers to entry to the payments industry, outside of banking, are now higher if you’re new to the payments market. This is causing some of the fintechs to approach the banks and propose collaboration,” says Gardner.
“It makes a lot of sense if you’re a fintech looking to enter the payments space to join up with a financial or a banking partner that knows the local regulations and markets. Putting those two together creates quite a powerful force,” he adds.
Open banking regulation is also spurring payments and AI technology innovation. Because of open banking regulations, Gardner says: “Corporates are starting to put their receivables factories on top of the payment factories and centralising receivables for better STR, for better visibility, better cash flow forecasting.” A payment or receivables factory being the centralisation of the payments or receivables process within an organisation.