Open banking has spent a decade laying down data highways: APIs, consent frameworks, secure sharing. Now agentic AI shows up as the autonomous car. It can reason, plan, call multiple systems, and take action with minimal human input.
In the words of Datos Insights, banks that master both data access and autonomous action “will define the next decade of financial services.” And it’s all converging right now.
The pressure cooker retail banks are already in
Retail banking economics are tightening fast. Boston Consulting Group estimates many incumbent banks are stuck with cost-to-income ratios of 60% or more, while well-run digital banks operate closer to 35%. Revenue growth in retail banking is expected to slow to just 2–4% a year through 2029.
At the same time, customers are more willing than ever to walk. J.D. Power finds that 52% of U.S. retail banking customers are open to switching banks in the next 12 months.
Direct banks and fintechs are leaning into that softness. They’ve removed friction from account opening, raised savings rates, and made it almost effortless to move balances. Direct banks now outperform traditional banks on satisfaction by 24–35 points in checking and nearly 90–100 points in savings based on J.D. Power’s 1,000 point scale.
That’s the backdrop for open banking and agentic AI: one makes it easier for money and data to move; the other will decide where it moves and who captures the economics.
Open banking paved the roads. Agentic AI decides who owns them
In our recent webinar, “Open Banking x Agentic AI,” Ron van Wezel, strategic advisor in retail banking and payments for Datos Insights, laid out the issue clearly.
Many banks treated open banking as a regulatory requirement, not a competitive weapon. They built APIs and consent flows, then stopped short of building new revenue streams or defensible differentiation from competitors. Data flows, but very little actually happens with it.
Agentic AI changes what you can do with that investment. Datos defines agentic systems as AI that can reason, plan, adapt to new data and goals, and orchestrate multiple tools for end-to-end outcomes. Their research show agents can offer:
- Proactive retention agents that spot withdrawal patterns and intervene before deposits leave
- Spend-analysis agents that protect card economics with targeted rewards and cash-back
- Autonomous journeys for onboarding, KYC/AML, and fraud detection
This is where external research and practice line up. McKinsey and Company estimates that agentic AI could reduce bank operational costs by 20% or more and cut manual workloads by 30–50% in targeted domains.
But if banks don’t reinvent their process or how they serve, global profit pools, roughly $1.2 trillion, could shrink by up to 10% over the next decade.
In other words, open banking provided everyone with similar roads. Agentic AI is the autonomous vehicle that can navigate to its destination with minimal human intervention.
Reframing fundamentals into AI priorities
Van Wetzel made a point that should resonate with any COO: “I think boring needs to be sexy with AI.”
“Boring” in this context is not trivial. It is the heavy, regulated, often frustrating work that determines whether customers stay, regulators are satisfied, and fraud losses are contained, like:
- Onboarding and account opening
- KYC/AML and sanctions screening
- Card dispute and chargeback handling
- Real-time fraud and risk monitoring
Van Wetzel’s advice to banks is simple: “Pick your most painful, most regulated process” and redesign it for agentic AI from the start. That is where the automation multiplier shows up fastest, because you already have rich data flowing through open banking enabled processes.
Imagine an orchestrated journey where agents continually decide:
- Is this withdrawal pattern a precursor to churn?
- Does this cross-border payment match expected behavior across all accounts and external data?
- Should this SME customer’s credit line adjust automatically based on open-banking cash-flow data?
That’s why orchestration, not point solutions, becomes the foundation.
Where agentic AI actually fits
Zishan Ali Khan, senior solutions engineer at Camunda, brought a pragmatic lens to the hype. Not every problem deserves an agent. He put it this way in the session: “You have to think, do you want this to be deterministic, highly guardrailed, highly regulated? And do you already know what’s supposed to happen? And is it repeated every time, all the time? In that case, maybe AI becomes more of a functional benefit… So maybe agentic AI is the right solution, but sometimes it is not.”
If the process branches wildly based on context, involves multiple tools, and can’t be fully designed at build time, that’s where agents shine.
Ali Khan pointed to onboarding as the archetype. For a new-to-bank customer versus an existing one, the number of checks, data sources, and tools can vary dramatically. In his words, “when you have that level of deviation where you can use three tools versus 10 tools, that’s where agentic AI is a better fit.”
That’s also where process orchestration matters most. You need a layer that can:
- Coordinate legacy systems, open-banking APIs, and new agentic services
- Enforce guardrails and approvals for high-risk steps, controlling the level of autonomy an agent has
- Capture full audit trails for regulators, auditors, and risk teams
- Allow you to redesign the journey without rewriting everything underneath
McKinsey’s latest Global Banking Review calls this the move from “heft” to “precision,” focusing technology on the few journeys and domains where you can generate outsize value, not spreading AI across dozens of disconnected experiments.
Three moves to start now
van Wezel closes their research with three actions. Reframed for today’s banking leaders, these actions become:
1. Treat agentic AI as an orchestration problem, not a tools problem
Don’t buy ten disconnected point solutions. Invest in an orchestration layer that can bridge agentic AI with your open banking infrastructure and legacy systems, so AI agents can take action across customer journeys.
2. Redesign the work, not just the workflow
Don’t slap AI on top of a 20-step manual process. Collapse it into a handful of high-judgment decisions, and decide which ones belong to rules, which to agents, and which stay with humans.
3. Start where the stakes are highest and the data is richest
Onboarding, KYC/AML, fraud, and money movement are ideal early domains. They are heavily regulated, typically ripe for innovation as they’re critical to business, and as a result, central to deposit retention and customer trust.
The strategic question is no longer whether open banking and agentic AI will reshape retail banking. They already are. The question is whether your bank will be the one redefining the next decade or the one lagging behind.
If you want to see how leading banks and analysts are thinking about this convergence in detail, including live examples across retail banking and payments, watch the full webinar recording.
Start the discussion at forum.camunda.io