What role do you expect players like networks, acquirers, and ISVs to play as AI becomes more embedded in payments?
In the build versus buy or partner analysis, I think many of the incumbents you mentioned will buy or partner in at least some areas. Let me explain why. Since ChatGPT launched in November 2022, we have seen that virtually all the profits in GenAI accrue to the frontier. Customers always want the smartest intelligence, or the optimal intelligence at a given cost, and what was at the frontier 18 months ago you can’t even give away today. It was amazing then, and now it’s worth near zero.
That’s how quickly things are moving, and the pace of model progress is only accelerating. Now, new model releases are so rapid that the top model only remains #1 for an average of 35 days. In that environment, are you able to stay at the edge of innovation? We partner with amazing, highly capable companies, but they’ve determined it’s a very different skill set to systematically evaluate the 149 major models released per year and match the right model to every task, to say nothing of the work required to turn models—which are really just the raw material of intelligence—into a purpose‑built solution for payments. So, we’re seeing tremendous demand to partner, and we’re investing in tools helping our partners to embed our AI inside their own applications.
Looking ahead three to five years, what trend or shift in payments do you think is being underestimated right now and that players should be paying more attention to?
My answer is vertical AI. A moment ago, I said the base model is the raw material of intelligence. It’s easy to underestimate everything you have to build on top—the workflows, the systems integration, the use of proprietary data—that turns the model into an enterprise‑grade product that’s ready to buy and implement and solves for specific needs inside an industry. That’s what I mean by vertical AI.
And we can quantify this. Out of the box, OpenAI’s GPT‑5.5—which is a fantastic model—is only 55% accurate for chargeback responses in our Benchmark. However, when you pair GPT‑5.5 with the Findustry AI harness, which includes our expert prompts, few‑shot examples, and the outputs of our proprietary tools, its score increases to 96.7%. That’s an increase of almost 42% in absolute terms. What drives that improvement? AI is fundamentally dependent on data, so it’s either “garbage in, garbage out” or “gold in, gold out.” Horizontal AI systems are trained on the open internet, including blog posts and tweets, but in vertical AI we have the opportunity to partner with payment processors and human chargeback experts, which adds invaluable domain depth to our product.