Although artificial intelligence has been one of the hot topics of the past year, it has been around for a very long time, dating to the 1940s. Over the past 20 to 30 years, there have been many permutations of what is often called traditional AI—and not just in the world of finance. Now, we’re living through a new boom of the technology in the form of generative AI.
Billtrust, a B2B order-to-cash and digital payments software leader based in New Jersey, has been at the forefront of these emerging AI solutions for all types of companies. In a recent PaymentsJournal podcast, Ahsan Shah, Senior Vice President of Data Analytics at Billtrust, sat down with Christopher Miller, Lead Analyst of Emerging Payments at Javelin Strategy & Research, to talk about where AI is making the most significant impact on businesses, now and into the future.
What’s New in Generative AI
Industries have been using AI tools like anomaly detection and classification for decades. But the new wave of generative AI uses language models as a fundamentally different approach, which can then be combined with traditional AI to solve business problems. That has resulted in a lot of excitement but also a lot of confusion about what is and what is not AI.
Generative AI is a language-based capability, offering an interface that nobody expected to happen this quickly in the AI evolution. In the past, when individuals spoke in a common language, it required a great deal of modeling and training data. But with open AI, these foundational models can translate language to code, to an action language, or to SQL. “It doesn’t circumvent what could be done with traditional AI,” Shah said, “but now you have a different toolkit in the box to be able to use language in capabilities, and fundamentally all of our customers speak in certain languages. And so this offers almost a new door of possibilities and features across the stack, across the industries, across various domains.”
Miller added: “What you’re saying is we shouldn’t really talk about artificial intelligence, but maybe about artificial intelligences. The ways that we would use these different flavors of artificial intelligence really vary from one another. The types of problems or business applications that are right for generative AI might not be right for a strictly deep learning approach.”
Where generative AI may be most helpful is in the interface layer with customer service, to help with natural language questions about data. “When I think about generative AI, it is an augmented assistant pattern to yourself as an individual, whether you’re in marketing or sales or at a SaaS (software as a service) company, whether it’s B2B or B2C,” Shah said. “It’s almost like an add-on assistant. A situation that might have taken a collections agent a long time—to find out which buyer to go to and personalize those types of workflows—is going to have a much more efficient process using AI, with the human element augmenting it to reduce that overhead.”
Moving Beyond the Chatbot
To this point, many people’s interfaces with AI have been limited to a chatbot that pops up when they access a website.
“There’s a difference between a chatbot that is consumer-facing and is the only thing that the customer interacts with and an interface that is used by a customer service agent to retrieve information who delivers that information through a chat interface,” Miller said. “With new technologies, folks often start by trying to match the technological capability to, for example, the type of data that’s available. You might further then segment your customer base and determine which type of experience they’re going to have, and different technologies may be appropriate to deliver those different experiences.”
Companies can personalize the experience far more than what was possible even a year ago. The chatbot can be prompted to say here’s the person you’re speaking to and here’s their background, and even suggest the tone of communication. “You can tell it to be soft with the person, because they’re frustrated after waiting at an airport for five hours with their family,” Shah said. “That’s information that’s out there, but your system has to be designed in such a way as to have that information source as part of the dataset. We want to talk to customers, understand their pain points, then use a phased approach in embedding AI workflows into our products in a systematic way.”
Enterprise value creation from generative AI is a different engineering exercise from a simple chatbot, requiring enterprise data far beyond the window of what Open AI or ChatGPT can do. “What I recommend people to do is start small, with use cases that have tangible business value, and really go out there and explore,” Shah said. “The worst thing you can do is to do nothing.”
A simple thing like invoicing can be a helpful place to start. “When we look at our customers, their goal is to get paid faster,” he said. “That’s really in a nutshell what we try to do. When an invoice comes in, it has to get paid, and if it doesn’t get paid, it goes to collections. You can look at something like that and ask, ‘What if I knew using traditional AI that an invoice that might be forecasted to default?’ But I might know something about the buyer based on their communication, their emails, and the correspondence to feed a personalized recommendation to that buyer. It combines personalization content with enterprise data.”
The Next Step in Accounts Receivable
Billtrust is building a unified accounts receivable system that can converge these functions to optimize the cache flows. The notion that the back-and-forth negotiating can actually take place in real time and perhaps even at the point of a transaction creates some very interesting potential opportunities. If, for example, a company’s agent and a seller’s agent are able to negotiate in real time, it’s also possible that financing offers could be included and evaluated in something near to real time. As the process becomes more templated, early adopters are likely to get an advantage.
“If you’re looking for something that’s extremely atomic from a financial transaction perspective, you may not want a generative AI model going and actually processing a payment for you,” Shah said. “Where you do want it is in that interface layer that helps with customer service that helps with natural language questions on your data. There’s no doubt the different types of AI will eventually help you reduce manual overhead, simplify workflows, and ultimately deliver a better user experience.”
