The Future of Finance Depends on Better Workflows, Not Better AI

Artificial intelligence is not creating a finance revolution as much as it is conducting an audit.

John Landy, chief technology officer at Billtrust, told PYMNTS that the lesson chief financial officers should carry forward from previous waves of digital transformation is straightforward. AI cannot compensate for poor enterprise architecture.

“The biggest issue is around having something be a bolt-on investment after the fact when you have fragmented data and solutions and systems and vendors working in your current environment today,” Landy said. “If you think about re-architecting from the data all the way up through and think about control and context throughout each of those, then you can really reap the most rewards from the AI investment.”

The observation reflects a broader shift occurring across enterprise finance. Competitive differentiation is moving away from purchasing AI capabilities and toward designing the operating environment that allows those capabilities to function effectively.

AI Is Only as Good as the Financial Context Behind It

The accessibility of AI can obscure its dependency on the systems beneath it. Powerful models are broadly available. Accurate, connected and operationally useful data is not.

“The AI tooling is available and the AI is available to all end users at most organizations,” Landy said. “It will present the data however accurate you give it.”

Accounts receivable makes the problem especially visible. The function often spans multiple ERPs, payment systems, customer records, vendor applications and manually maintained spreadsheets. Some systems update in real time. Others operate in batches. Employees reconcile the differences.

“Most organizations are dealing with complex environments. Multiple ERPs is the standard sort of system background, as well as many vendors … large and small vendors across the ecosystem to get their work done,” Landy said, noting that spreadsheets also add another layer of temporal uncertainty.

“You can think of a spreadsheet as a snapshot of data in time that you are trying to correlate across real-time systems, batch systems and reconciliation processes,” he said.

Installing artificial intelligence above that environment does not resolve the discrepancies. It gives the system a faster way to process them.

“AI has to have visibility into all of it,” Landy said. “It cannot be added after the fact.”

That principle is starting to show up in how AR data reaches AI tools. Rather than requiring finance teams to pull data into a separate platform, live invoice-to-cash information can now be exposed directly inside the assistants people already use for decision-making, asking a plain-language question and getting a data-backed answer without opening a report or switching tools.

It’s a change in interface, but it reflects the same underlying idea Landy described where the value isn’t in adding an AI layer on top; it’s in placing accurate, structured context where the question is actually being asked.

The Real Productivity Gain Is Time to Think

Much of the AI conversation in finance focuses on headcount, task automation and processing speed. Landy points to a more consequential gain: reducing the time finance teams spend assembling information before they can make a decision.

“If you can save the time people spend assembling information, that is the biggest difference,” he said. “They can spend that time thinking and making better business decisions.”

That shifts the finance function away from producing retrospective reports and toward managing events as they develop.

A late payment from a consistently reliable customer, for example, may appear insignificant in isolation. Connected AI systems can recognize that the payment has broken an established pattern, assess its impact on a cash forecast and alert the appropriate employee before the missed payment becomes a collections problem.

“If you can detect in real time that a payment has not arrived when it typically does, you can react and have a more meaningful conversation before it becomes a problem,” Landy said.

The distinction is subtle but important. AI’s value is not merely that it helps finance teams respond faster. It allows them to intervene earlier, when more options remain available.

Autonomous Finance Needs Boundaries

Not every workflow should move toward full automation at the same speed.

Cash application, invoice matching, aging analysis and fraud flagging are natural candidates because they involve high-volume processes, recognizable patterns and measurable outcomes. The role of the human can be limited to reviewing exceptions.

The calculation changes when a decision involves a strategic customer, a large dispute or a meaningful legal or reputational risk.

“If you have an important relationship, you are not going to want the process automated to the point where no one is available to handle the communication,” Landy said.

The question for CFOs is therefore not whether humans remain involved, but where they sit in the system. Landy separates workflows into those where employees are “in the loop,” those where they remain “on the loop” as supervisors and those that can operate autonomously.

That governance cannot be applied after deployment. It must shape the workflow from the start.

“Anything involving legal, brand or reputational risk should have a human in the loop,” Landy said.

The CFO Becomes a Context Architect

Artificial intelligence also changes the traditional division of labor between finance and IT. CFOs do not need to become machine-learning engineers, but they do need to understand how financial data is stored, shared, secured and introduced into AI systems.

“The number one requirement is a partnership between CFOs and their IT teams to ensure the right infrastructure is being evaluated and addressed,” Landy said.

Over time, that partnership may evolve into a broader role for finance leaders.

“Every business unit owner, including finance leaders, should move toward becoming a context architect for their solutions,” Landy said, “working with human and digital coworkers to deliver meaningful results.”

Watch the full PYMNTS TV “Summer School” interview with Billtrust CTO John Landy to hear more about:

• Why AI cannot repair fragmented finance infrastructure. Landy said bolt-on deployments will struggle when accounts receivable data remains scattered across multiple ERPs, vendor systems, batch processes and spreadsheets.
• How connected AI can move finance from reporting problems to preventing them. The discussion explores how real-time visibility into payment behavior can help teams identify late-payment risks earlier, improve cash forecasting and spend less time assembling information.
• Why autonomous finance still requires deliberate human boundaries. Landy argues that high-volume tasks such as cash application, aging analysis and fraud flagging are strong automation candidates, while credit decisions, major disputes and legal or reputational risks should retain human oversight.