Visa Ventures: This is how AI will transform the future of payments

===

Arranging a holiday is exciting, but stressful. Having to book flights and accommodation is never fun, but what if you could tell a programme roughly what you want, and it goes out into the world and does it all for you? That’s the kind of application that payment processing giant Visa wants to support with the $100m generative AI fund it launched last year.

It’s been three decades since Visa started using artificial intelligence – though obviously nowhere as powerful as models are today – for things like risk and fraud management. That range is now much wider.

“Today, we’ve got several hundred AI models in production across different parts of our businesses, working to solve these real problems for banks and merchants and consumers. The impact of these are enormous,” says David Rolf, head of Visa Ventures, on a recent episode of the CVC Unplugged podcast.

The new generative AI fund, which was announced in October last year, will be slightly more flexible in its investment criteria than the rest of the unit. Visa Ventures typically invests in series A to C funding rounds. The AI fund can invest at earlier stages to have more of an impact on startups with nascent technology.

“We really see this as a transformational technology,” says Rolf. “We think there’s going to be a lot that changes.”

“The rationale for that announcement and setting up that initiative is we’re taking this extremely seriously inside our company. We’re using generative AI for some of our own products, and we’re extremely interested to see how it’s going to remake different facets of our industry.”

Whether it’s preventing fraudulent transactions in real time, managing data, having language models for use internally within Visa, the company has been all in for a while. Through the new fund, which is being operated under the auspices of Visa Ventures, they want to push AI usage further and, as Rolf puts it, take the friction out of a myriad financial functions.

“Talk to anybody in the intelligence or security community, and they have a lot of concerns about what deep fakes will mean for account opening, for fraudulent transactions, all this kind of stuff,” he says.

“Bad guys are going to be using it, so we need to understand what they might do with it.”

Some of the companies, says Rolf, could see meteoric growth – while the technology itself is new, the problems it looks to solve have been ubiquitous for a while. The opportunity is a big one.

Even something as deceptively simple as gleaning insights from unstructured data like voice, video – which can’t be inputted into a basic SQL database – is actually incredibly difficult at scale, but a gen AI engine could, in theory, make that far easier.

AI agents

One of the more exciting applications is that of AI agents acting for individuals or business in the real world.

On the simple side, he says, that could just be telling an AI programme to go and find you options of toys you can buy your newborn, but it can get much more complicated than that.

“I’m interested to fly to a major European city this summer with my wife and 2 sons. Please put together a sample itinerary and bring me back a couple of good flights,” says Rolf as an example. “That’s really complex. Choose the city – am I choosing the city first or am I choosing the cost of the flights?”

All of a sudden, the financial stakes entrusted to a piece of software get very high, very fast. More complexity comes in when you consider there may be a real person on the other side of the transaction, or if there are any adjustments or nuances that need to be considered.

Payment hassles

The hope is that AI will help Visa address a wide set of common problems.

If you live in a developed economy you might take most payment technology for granted, but beneath the surface, the difficulties and inefficiencies are legion. Business-to-business, real-time, and cross-border payments all come with own set of challenges that compound.

Even something as commonplace as invoicing remains a huge point of friction. Determining if the payment terms are up to date, if the itemisation is correct, if the payment windows are accurate, if invoice number matches the right business account, these are just some of the issues take up a large amount of a business’ time.

“The hard part about business-to-business payments is not actually the money movement, it’s the right amount of money and it’s the reconciliation,” says Rolf.

“If I’m buying merchandise and some of it got on a ship leaving Vietnam on June 30th, but some of it didn’t ship until July 15^th, and my contracted rate changes at the middle of the year – as complex a situation as you can imagine happens quite frequently.”

Things get more complicated still when these transactions are done across borders and you have to deal with changes in foreign exchange, different regulatory environments and transparency in different countries.

“You can have one invoice that could be for 17 different shipments.  So the data is really complex in business-to-business payments. It’s still really bad for large companies. It’s still really bad for small companies. It’s still really bad for medium-sized companies.”

Fraud prevention, naturally, is also a constant area of focus, not least because it’s never static – the more progress you make, the more progress they make on the other side.

“You build a better mousetrap and the fraudsters figure out how to get around the mousetrap, right? So that’s continually an area we’re interested in.”

Corporate AI funds: The list