Payments modernization is accelerating, but the way banks deliver these programs remains stubbornly manual. Oliver St Clair Stannard, Head of Go to Market at RedCompass Labs, argues that AI has the potential to change that, but until its impact can be measured and proven, it will struggle to move beyond experimentation.
Canada is in the midst of the most significant overhaul of its payments infrastructure in a generation.
The Real Time Rail completed system integration testing at the end of 2025 and has now moved into user acceptance testing, with new participants onboarding. Open banking is moving from policy to implementation, and scheme deadlines are no longer theoretical.
This is not unique to Canada. Across North America, payments modernization is accelerating. In the United States, FedNow has rapidly expanded participation, while real-time payment volumes continue to rise sharply.
The Clearing House’s RTP network recently processed more than 1.8 million transactions worth $5.2 billion in a single day, and real-time payment volumes across North America are projected to reach over 8 billion transactions in 2026.
The scale of change is clear, as is the volume of work required to deliver it. What is less often discussed is how that work actually gets done.
The most manual part of banking is the one that builds it
For all the investment in infrastructure, the delivery lifecycle itself remains highly manual and particularly labor-intensive. Banks have sophisticated tooling, but moving from requirements to tested, production-ready systems still depends heavily on human effort.
Each stage introduces friction, with requirements interpreted and reinterpreted. Test coverage does not always map cleanly back to the original intent. Reviews, rework and reconciliation consume significant time and effort.
This is where transformation programs slow down. It is also where costs accumulate and risk increases.
The industry focuses heavily on what is being built, and far less on how it is delivered. Yet that is where many of the biggest constraints now sit.
AI has potential here but lacks credibility
AI is already well embedded across banking. It supports fraud detection, credit decisioning and customer experience. According to KPMG Canada’s 2025 GenAI survey, more than 90 per cent of financial services leaders see it as critical to competitive advantage.
But in payments, the conversation about AI is still largely focused on products and customer facing services.
The delivery lifecycle has received far less attention, even though it offers one of the clearest opportunities for impact. AI can already support requirements analysis, test generation, configuration validation and code development. In some cases, it is starting to do so.
Even where AI is being applied to delivery, there is still no widely accepted way to measure its contribution.
How much of a program has been delivered by AI rather than manual effort? How much time has actually been saved, or rework avoided? In most organizations, there is no clear answer.
Without that, it is difficult to build a defensible business case, particularly as transformation budgets come under increasing pressure.
The next phase will be defined by proof
That position is becoming harder to sustain. Payments programs are under increasing pressure as timelines tighten and expectations rise.
RTR timelines in Canada, FedNow expansion in the US, and ISO 20022 deadlines globally are all tightening the window for doing things the old way. Programs need to move faster, at lower cost and with fewer defects. The traditional delivery model is struggling to keep up with the scale and pace of change.
AI has the potential to address this. But potential alone is no longer enough. The next phase of adoption will depend on whether its contribution can be made visible and measurable. That means embedding AI into delivery workflows in a way that is transparent and auditable. It means developing clear ways to quantify its role in producing outputs and improving outcomes.
Without proof, AI will remain on the margins of delivery, rather than shaping how transformation is executed.
