Where will AI take data analytics? The possibilities are endless

Podcast: Play in new window | Download

Organizations have long been challenged with deriving insights from their data. Some have the capabilities and resources to do this, while others are far behind. Artificial Intelligence (AI) has the power to leapfrog data analytics into the future, embedding enterprise analytics into the day-to-day overall health and success of a company.

Billtrust has been at the forefront of building out analytics processes using AI, particularly in the payments space. In a recent PaymentsJournal podcast, Ahsan Shah, SVP of Data Analytics at Billtrust, spoke with Christopher Miller, Principal Analyst, Emerging Payments at Javelin Strategy & Research, about the future of AI-powered data analytics.

Democratizing AI

That’s not to say that organizations aren’t considering AI anymore: Success for most organizations will come from the democratization of generative AI, not top-down mandates.

“Some companies are further along than others just by giving their employees the opportunity to experiment in the form of goals and self-training,” Shah says. “Some teams at Buildtrust are even doing hackathons to learn how to do this cool stuff. I think this is a natural progression, and I think that's the right way to go.”

AI is poised to move from a world of basic models to a large set of tools, instruments, infrastructure, and services. Technology advances are moving much faster than adoption. OpenAI is already at the forefront of multimodality.

“There's been an explosion in the number of different systems that oversee different parts of how a business runs, from the front-line customer interactions to the granularity of the actual payment processing and chargeback process, all the way down to the timing of revenue recognition and how you manage cash,” Miller says. “One of the challenges for the team is figuring out how to make all those different pieces fit together.”

Data Explosion

Most businesses have someone compile various pieces of information or cut and paste data into spreadsheets. They may have a dashboard that compiles various pieces of information, but it can be difficult to maintain that dashboard and add new data as it comes out. The explosion of data creates opportunities for insights, but it also creates challenges in terms of scale, especially for organizations with limited teams and resources.

This idea of ​​cross-functional analytics is challenged not just by the volume of data, but also by its structure. “There are three different vectors at play here,” Shah says. “The sheer volume of data, the urgency to act on it, and the proliferation of different functions. Companies need better ways to integrate data across functions and get it to the right people who can act on it, and this is something that is often overlooked.”

Emerging generative AI technologies could offer one way to solve some of these problems, including new ways of creating reports, where instead of simply handing definitions off to engineering teams who create the reports, data isn't pushed out of the system, but rather pulled from the system by people who are in a position to act on its insights.

The new term is generative BI (generative business intelligence) – simply asking specific questions in human language, such as “What anomalies do you see in the payment patterns of buyers on the West Coast?” – something that would traditionally take weeks of engineering analysis.

“It's a rapidly growing field,” Shah says. “Six months ago, there might have been one or two companies on the market with an LLM product that we could use. Everyone was writing poetry on ChatGPT and experiencing the power of language models firsthand. But most people were also running headfirst into the challenges of the data collection side of those models, which provides the interaction layer and doesn't necessarily provide insights. That's the next step.”

Beyond ChatGPT

ChatGPT users are limited to a context window: they can type questions, but the tool has no knowledge of the user, their enterprise data, their CRM, or their transactions. Integrating the data and analytics layers directly into LLM requires model engineering and domain fine-tuning.

There's only so much you can do with a basic model. How do you make your data open, extensible, and engineered so that generative AI can get the most out of it? This is something we at Billtrust are actively working on.

“We're currently preparing to launch our Copilot product, essentially incorporating an enterprise-grade secure interface like ChatGPT,” Shah says. “Instead of going back to the traditional way of hiring a data analyst to create reports for you, you'll go into Copilot and ask specific questions. You should think of this as a way to improve your operations, rather than a complete transformation.”

Several companies are already rapidly leveraging this capability – Open AI, as well as Facebook Meta, AWS, and Claude Anthropic are integrating it – and you'll start to hear the term agent workflow.

“This seems very visionary, but I don't think it's that far off,” Shah says. “You'll see a world where people log into your SaaS product or your B2C product and simply ask, 'Book a trip for me and my family,' and it takes you through a multi-step flow to book a hotel. You can apply this to B2B right now. Instead of booking a trip, you can run campaigns and target these customers.”

The Need for Governance

When systems operate based on limited instructions from humans, interoperability of those systems becomes important, which suggests the need for standards and essentially another layer of API development.

“It's important to have governance to avoid problems and catastrophic impacts with AI,” Shah said, “but you can't do it in a way that hampers companies' ability to innovate and build great products.”

Another concern is cost: it is high and still rising. While unit costs are slowly starting to come down, absolute costs are increasing as the model exponentially adds more tokens, creating additional computing demands to support it.

But the possibilities far outweigh the challenges. “You're only limited by your imagination,” Shah says. “The best implementation at the agent level will create the largest world of creative freedom — one that gives artists the ability to focus on what they do best, eliminating friction and redundancy in other tasks. Long before the implementation is perfect, we'll have the technical capabilities to support that kind of imagination.”

“There will be a thorough knowledge of how to effectively use different models for different businesses. I think there will be an explosion of options. There may be a bit of chaos for a while until the dust settles.”