The Green Sheet Online Edition
August 25, 2025 • 25:08:02
GS Advisory Board
AI in action: What's working in payments and what's not
In our first August issue, advisory board members shared how they're putting AI to work today, from fraud prevention to underwriting and beyond. In this installment, the conversation widens. Members highlight what's working, where AI initiatives fall short and how its rapid evolution may upend traditional ISO models.
We wish to thank Benjamin Abel, Mark Dunn and Dee Karawadra for taking time to share their perspectives herein. And we appreciate all advisory board members who participated in this Q&A. Below are the questions, followed by their responses:
- Have you implemented AI in your business? If so, what have you learned—successes, challenges or unexpected outcomes?
- In your view, how is the broader payments and fintech ecosystem leveraging AI effectively today? Are there any high-profile efforts that haven't lived up to expectations?
- Looking ahead, how do you see your business, and the industry at large, evolving with AI in the mix? Please share any specific trends, innovations or concerns you anticipate.
Benjamin Abel, BAMS
AI, in many industries not just ours, is definitely the shiny new tool in the toolbox with everyone trying to find a use. I think we have seen some companies effectively leverage it in fraud prevention, underwriting and optimizing payments, but not all rollouts have gone as smoothly as hoped. Visa's Protect suite, which includes Visa Risk Manager (VRM), Visa Advanced Authorization (VAA) and Visa Deep Authorization (VDA), claims to use AI to effectively help with risk scoring, which prevents fraud and gets more legitimate transactions approved.
There are also lenders like Upstart that are pulling in alternative data to approve more borrowers without having to expand their overall risk tolerance. And you have Stripe, which uses what they call "intelligent payment routing," where they claim their AI can predict which acquiring banks, card networks or processing paths are most likely to approve a given transaction based on factors like card type, issuing bank behavior, geographic region, currency, time of day and even past declines.
We can dig into the data to see how effective each truly is, but they are practical, high-impact uses where AI looks to really shine because the data is strong and the goals are clear.
That said, not every AI initiative has hit the mark. Take the Apple Card situation in 2019. Goldman caught heat when it was claimed their AI underwriting gave lower limits to women, after developer David Heinemeier Hansson alleged he received 20 times his wife's credit limit even though she had a higher credit score than he did.
This became a massive PR headache and raised concerns about whether "gender blind" algorithms (or you can replace "gender" for other identifying characteristics in other applications) could still potentially produce biased outcomes through proxy variables. The New York State Regulator concluded in March 2021 that there was no clear evidence of unlawful discrimination against women, but the reputational impact had already been done.
Looking ahead, I think AI is going to force ISOs to evolve faster, creating more differentiation. We're already seeing smarter fraud tools, better transaction routing and predictive underwriting models—but what's coming next can be more transformative.
I expect AI to reshape agent support, pricing analysis, portfolio risk modeling and even merchant onboarding. The ISOs that learn to integrate AI into their workflows, rather than waiting for upstream partners to do it for them, will be the ones that differentiate in the market, streamline their overhead and scale faster.
That said, we have to be mindful of the bigger picture. There's a real risk of many smaller ISOs being cut out of the equation if they don't evolve fast enough to demonstrate that they offer any true benefit.
Acquirers and processors are investing heavily in AI to automate more of the value chain, and some may eventually see less need for traditional ISOs for relationship building and direct to merchant marketing. So while AI creates opportunity, it may also force us to have to rethink our positioning.
Something that's always been true, though, is that the future belongs to those who could add value beyond just selling accounts, whether that's through smarter tools, consultative service or vertical-specific expertise. Anyone who recognizes AI as one more tool in the kit, and can effectively leverage it to give themselves a competitive advantage, will end up enjoying the success of their new Model T while they leave behind everyone who's still shoveling horse dung.
Mark Dunn, Field Guide Enterprises LLC
I see the traditional ISO business using AI in traditional ways: to market better, compose more attractive sales materials, sell more effectively, install quicker and support merchants better. What I am not seeing—and am concerned about—is the traditional ISO industry realizing the potential for powerful AI companies to completely reformat the payments industry from a position of power and dominance.
Companies that use the power of training large language models with massive amounts of payments data can achieve better control of risk and fraud, accelerate the flow of funds to businesses, provide dashboards and reports that better inform business owners, make suggestions as to how to manage their businesses better, and provide a more convincing sales proposal than traditional ISOs. They can sell more effectively direct to merchants without the high cost of sales agents' residual share.
Our industry is a huge target for companies that can dominate with AI. We have recurring revenues, payment acceptance that is embedded in the merchant and cardholder communities, decent profit margins, and the opportunity to quietly raise prices from time to time. AI masters complexity.
The complexity of merchant services has been a rampart that kept many invaders away for a long time. I see this changing at a rapid pace.
Dee Karawadra, Impact Pays
We're in the thick of it. At ISOhub, we're actively training AI to analyze past merchant and agent notes to suggest resolutions—basically, a support assistant who never takes a day off or forgets what happened last month.
But we're not just rushing it out the door. We're still in the training phase, making sure it's fed the right data and giving back the right results. We won't release it until we're confident it actually helps, not just looks fancy in a sales deck.
Fraud detection and underwriting are clear wins. AI's making fast, smarter decisions. But there's also a lot of noise. Way too many folks are slapping "AI" on what's really just a glorified chatbot. If it's only answering FAQs and can't handle nuance, it's not innovation; it's marketing with a hoodie on.
For us, AI is about helping ISOs punch above their weight: automating residual insights, speeding up support and giving agents tools that used to require a full team. Industry-wide, AI will eliminate a lot of the grunt work, improve decision-making and (hopefully) improve the merchant experience.
My concern? That companies will get caught up in hype and launch half-baked solutions. We're taking the opposite approach: train it right, test it hard, and roll it out only when it works for real people in real workflows. 
Notice to readers: These are archived articles. Contact information, links and other details may be out of date. We regret any inconvenience.
