How AI Is Rewriting Acquiring

Artificial intelligence has long been a part of payments, known early on as machine learning, but now it’s shifting from a predictive function to agentic AI, where systems act, not just advise, said Joseph Arthur, managing partner at RPY Innovations, a payments consultancy.

Speaking on a panel at the Northeast Acquirers Association conference last week in Boston, Arthur said the era of AI watching and waiting for instructions, its advance heralds a shift. “Predictive AI suggests, agentic AI acts,” he said. “Imagine a system that doesn’t just flag a suspicious transaction, but autonomously initiates the dispute and adjusts accordingly.” He says this changes the math of margins, head count, and risk. Agentic AI employs machine-learning models that can mimic human decision making in real time, in IBM’s definition.

In the case of underwriting, today’s model is a snapshot in time of the merchant’s risk, Arthur said. How does AI affect underwriting when it can monitor continuously?

Having a decision made on one snapshot at a time doesn’t really get to the full picture, said Kham Inthirath, chief executive and founder of Compound Effect, a business-advisory firm specializing in AI. Human oversight is essential, especially when it comes to defining the guardrails for what the AI engine is allowed to do. An employee has to ensure the agent is behaving and providing the right types of outcomes and decisions, he said.

AI intervening in risk assessment also requires know-how to explain when an underwriting decision made or influenced by AI affects a merchant, said Bill Lynch, senior vice president of sales at Aurionpro Fintech, a professional services firm. “in other words, being able to explain to someone, in this case a merchant, why the situation has changed,” Lynch said.

If a transaction or transaction pattern triggers a potential adverse decision, that has to be explained so it’s not a surprise to the merchant, he said. Having policies and guidelines in place for these situations is critical. “But having that in place so that you’re not catching them completely off guard and being able to provide them real reasons, not just because that’s what our machine told us … is really important,” Lynch said.

That’s just one payments operation that could be touched by agentic AI. By 2027, panelists predicted, AI integrations will be table stakes for independent sales organizations. “Its ability to draw on complex and highly detailed and deep information, giving access to the data really quickly,” Inthirath says.  

“It’s going to be fraud prevention and not just fraud recognition,” says Colin Luce, chief executive and cofounder of Basis Theory, an agentic-commerce and payments-security provider. That may be out there and may not happen by 2027, but steps toward that will be something that ISOs can use, he said.