Ask a payments leader whether their company uses AI, and the answer is almost certainly yes. Ask whether it has changed the P&L, and the conversation becomes far more complicated.
The numbers reflect that tension. In McKinsey’s 2026 State of AI survey, nearly 90% of respondents report using AI regularly, yet only 37% attribute any EBIT impact to it. Adoption is no longer the challenge. Results are.
A large opportunity, with a catch
Few industries have as much to gain as payments. McKinsey estimates that generative and agentic AI could unlock up to $110 billion a year in operational productivity across the sector.
There is, however, a detail that tends to surprise executives. Even when AI removes more than 60% of operational tasks, net cost reductions typically land between 15% and 20%. The gap between those two figures explains why so many AI initiatives look promising in a pilot and disappointing in the budget. It also points to what the companies capturing the most value are doing differently, which is where DEUNA Playbook No. 1 begins.
Where AI actually fits in a payments operation
Rather than treating AI as a single initiative, the playbook breaks it down into three areas, each with its own challenges.
The first is payments intelligence, and it starts long before any model is deployed. When every processor, acquirer and payment method reports in its own way, even a simple question about last week’s approval rate can take days to answer. According to the Merchant Risk Council, 80% of merchant professionals struggle with at least one data or technology issue. The playbook explains why this is the point where AI initiatives in payments are often won or lost.
The second is performance optimization. Behind every payment is a chain of decisions: which payment methods to show, how to process the transaction, where to route it and whether to retry it. Many of those decisions still depend on rules written months ago, in a market that changes every day. The playbook shows how AI agents approach the same decisions one transaction at a time, and why a well-chosen retry can matter as much as the right route.
The third is process automation. Global chargebacks are forecast to reach 324 million a year by 2028, and the back office already absorbs much of that pressure. Which parts of that work can AI take on, and which should remain firmly in the hands of the team? The answer is more nuanced than “automate everything.”
The question to ask before giving AI autonomy
As AI moves from recommending to deciding, control becomes the defining issue. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, with inadequate risk controls among the reasons. The playbook outlines the four guardrails that successful payment implementations put in place from day one, including one that many teams overlook until a model starts behaving differently than expected.
Where does your operation stand?
The playbook closes with a short self-assessment designed to be completed with your team. Three questions are enough to show which of the three areas holds the greatest opportunity for your operation today.
