PaymentsJournal Podcast: Programmatic Payments and the New Compliance Challenge

When Software Initiates the Payment, Controls Have to Keep Up

A payment used to begin with a person deciding. But that decision is increasingly being handed to software. As agentic AI continues to grow, this trend will only continue. Ensuring these systems behave as intended and documenting this will be the challenge for banks and payments companies.

In a recent PaymentsJournal podcast, FinScan's Kieran Holland, Global Head of Solutions Engineering, and Chris Ostrowski, Head of Product Management, joined James Wester, Co-Head of Payments at Javelin Strategy & Research, to discuss where programmatic payments stand today and what they demand of compliance, monitoring, and oversight.

A more transactional world

Businesses and consumers now generate a steady stream of smaller transactions triggered by events, thresholds, and conditions. Holland describes programmatic payments as the background technology driving that shift. It covers everything from a subscription renewal to a voice assistant restocking groceries. Ostrowski compared the consumer version to a limit order in the stock market where the purchase executes on its own once the price is right, except the same logic now runs across a 24/7/365 economy. Consumers experience this as handing off a chore rather than giving up control.

Money moving at machine speed

For larger organizations, payments are being embedded directly into operational systems rather than handled as separate financial events.

That speed changes what compliance technology must do. Ostrowski shares these payments can't wait for analysts to arrive at eight in the morning, so the monitoring has to run around the clock and evaluate transactions as they flow through. The same applies to the checks that happen before a payment ever executes. Screening a counterparty or running due diligence on a newly introduced third-party agent has to work at the speed and scale of the business process it supports.

Controls built around human behavior meet machine behavior

Automated systems don't act like the people that most controls were designed around. As Wester put it, agents won't necessarily behave in ways we consider logical. They follow patterns recognized in data. Institutions will need to learn what normal activity looks like when transactions are machine-initiated, because the baselines they've built over years describe someone else.

The data underneath the decision

Holland pointed to studies of major corporate AI rollouts where the failure wasn't the model or the business objective. It was that the data going in wasn't good enough to produce the outcome the organization needed.

The same holds in compliance. An automated decision about a party to a payment is only as reliable as the party data behind it, including names, addresses, and ownership information. In a programmatic environment there's no analyst reading around a truncated name or a missing identifier, so the quality of that data determines the quality of every decision made on top of it.

People still make the final call

Programmatic payment systems won't reach their potential without people who can oversee performance, provide direction, and step in when judgment is required. Ostrowski's advice was to build that support as the program succeeds, rather than catching up after examiners arrive with findings. Holland's version was blunter: avoid the AI equivalent of throwing spaghetti at the wall to see what sticks.

Listen to the full conversation at PaymentsJournal.