Why AI in payments still has an enterprise execution problem

Artificial intelligence is rapidly reshaping the payments industry’s underlying architecture and redefining the user experience. Yet, while AI adoption is accelerating, the journey from promising pilots to scaled business outcomes remains far from straightforward.

Opus Chief Marketing Officer Nischala Murthy spoke with Chief Revenue Officer Prakash Devulapalli to explore one of the most important questions facing the industry today: why, despite significant investment and experimentation, do so many AI initiatives struggle to deliver measurable enterprise value?

The broader evidence suggests an industry making meaningful progress, but unevenly. Stanford’s AI Index 2026 notes that financial services stand out for their comparatively strong use of AI in core functions such as risk and compliance. However, success within individual functions does not automatically translate into enterprise-wide transformation.

That distinction sits at the heart of this conversation.

Prakash offers a grounded perspective on where real returns are emerging, why productivity gains are often overstated, and what banks and payments firms must do to prepare for a world increasingly shaped by agent-mediated interactions. His message is clear: winning with AI will require more than better models or isolated use cases. It will demand operating discipline, common standards, and a long-term view of how technology, processes, and business models must evolve together.

The next chapter of AI in payments will not be defined by who experiments first, but by who learns to execute at scale.

Nischala: Where are clients seeing the most meaningful AI opportunities today, and what is making it difficult to turn those opportunities into measurable value?

Prakash: On the technology side, the immediate focus is productivity. Organizations are using AI to improve the run processes across ticket management, production support, BPO, and infrastructure, to drive productivity and velocity into traditional processes.

On the business side, the conversation is shifting toward how AI can create new features and client experiences. Enterprise applications are becoming less UI-led and more AI-led; products are becoming increasingly headless, with agents interacting via protocols such as MCP (Model Context Protocol), a new standard for connecting AI assistants to applications like CRMs so they can read and act on live data. In that sense, many products are now being redesigned to be AI-ready.

Cybersecurity is another major area of opportunity. As organizations test newer models, they are uncovering vulnerabilities at a scale they had not previously identified, driving reprioritization across the enterprise.

The harder issue is measurement. The central question remains, “How do I evaluate ROI?” Overspending on tokens does not, by itself, prove value creation.

Nischala: With AI use expanding across these areas, how are clients thinking about ROI and value realization?

Prakash: Some technology companies and startups are already seeing tangible benefits. Code development, testing, optimization, and even prototyping have accelerated meaningfully. Adoption is clearly rising across both financial startups and enterprises.

That said, the clearest gains are still showing up in startups and among individual builders rather than at the Fortune 500 level. As a result, expectations have become more concrete, while large-scale value realization remains work in progress.

Nischala: If value realization at enterprise scale is still a work in progress, where are you seeing the most meaningful returns today?

Prakash: In financial services, some of the earliest production use cases are in customer engagement. There is also less reliance on traditional BI dashboards as users become more comfortable querying enterprise data directly and receiving answers in natural language.

There are also early benefits in IT and infrastructure management, CI/CD deployment, and selected user-interface improvements. Co-pilots for back-office and human-driven processes are gaining traction, especially in operations. But operations remain in their early stages: in many institutions, the economic case is compelling only when AI translates into direct reductions in headcount. Elsewhere, the benefit is more about speed and operating velocity than immediate P&L impact.

Nischala: Given that these gains are still uneven across the enterprise, what misconceptions about AI adoption do you see most often?

Prakash: A common misconception is that simply making AI tools available will automatically make people more productive or reduce headcount. Tools alone do not solve the problem. Enterprise adoption requires guardrails, operating discipline, and a practical model for how people use these systems.

In most organizations, a small group uses the tools very effectively, while the majority does not change behavior enough to move enterprise outcomes. It is not the 20 percent who double productivity that changes the trajectory; it is whether the remaining 80 percent can improve by 10 to 20 percent.

There is also too much short-term thinking. Many leaders overestimate what can be achieved in the next three or six months and underestimate what can be built over the next two or three years. Given how quickly technology is evolving, enterprises need an operating system that can keep pace, rather than isolated pockets of excellence that create more tech debt.

Nischala: If enterprise adoption requires that kind of operating discipline, how should banks approach AI as the technology continues to evolve?

Prakash: Banking systems were designed to protect against non-human behavior. The architecture assumes that a human initiates the transaction, and controls are built around that assumption. That design premise will now have to change as agents increasingly mediate interactions.

Imagine ERP platforms become agent-enabled, banks will need to support machine-mediated engagement with clients and partners. At the same time, multiple protocols are emerging for how agents will interact in payments and commerce. Over the next three years, some standards will begin to consolidate, though they are unlikely to be uniform across regions.

For banks, the near-term priorities are clear. First, standardize AI adoption internally so productivity gains can scale across functions. Second, prepare the institution for agent-to-agent and machine-to-machine interactions across the ecosystem. Throughout, explainability will remain essential as regulated institutions must ensure that what they launch is understandable from a regulatory standpoint.

Nischala: As banks prepare for more agent-to-agent and machine-to-machine interactions, how important will industry standards become?

Prakash: In regulated sectors such as banking, standardization is essential. Without it, the industry will struggle to agree on how institutions and counterparties should interact.

But standardization should not be forced too early. The next phase requires experimentation and competition among technology providers. Over time, the approaches that prove most convincing will gain adoption, as they have in every major technology shift. That consolidation is likely to begin within the next two to three years.

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – –

Prakash Devulapalli is Chief Revenue Officer at Opus Technologies, where he leads Sales, Marketing and GTM. He brings more than two decades of experience across banking & payments, and financial services.