What banking leaders say when you ask them about AI challenges
Ask a banking leader what's blocking AI at scale and you rarely hear "we don't understand AI." Chris Cheyenne said as much on the BankingReinvented podcast: "AI is not new for banking. Banks, they are familiar with the use cases of AI, like for example, on the risk - model risk management frameworks, how you manage your risk, how do you model it, how do you split test it, how do you cohort them." The literacy problem most vendors diagnose simply isn't the problem practitioners describe.
What leaders flag is different. Banks have run AI in risk modeling for years. The difficulty surfaces when they try to move AI into frontline execution - into the work a relationship manager does across five disconnected tools, or the process a service agent navigates through four systems to answer one customer question. That's where pilots stall, not because the models are wrong, but because the operating environment the models are dropped into is structurally fragmented.
That fragmentation is the real diagnosis. When AI agents operate without unified context, consistent rules, or a single source of truth, they don't automate the frontline. They accelerate its inconsistency. The challenge banking leaders need to name isn't how to pilot AI, it's how to build the infrastructure that gives AI something coherent to act on.
Why 50% of frontline work lives in space no system owns
Most banks assume their frontline operates inside defined systems, but it doesn't. Roughly half of what frontline staff do every day - handling exceptions, chasing approvals, coordinating handoffs between teams - happens in the whitespace between those systems. No core platform owns it, no CRM tracks it, and it's held together by email threads, spreadsheets, and institutional memory.
This isn't a recent problem. Banks have accumulated disconnected tools across decades, and each one handles a slice of the customer journey. But the joins between those slices - the moment a mortgage application needs a credit decision that needs a compliance check that needs a manager sign-off - belong to nobody. Frontline staff fill that space manually, every time, and that whitespace is the real operating environment.
Now consider what happens when you drop an AI agent into that environment. The agent needs complete context to act correctly, consistent rules to follow, and a single system to write its output back to. In a fragmented frontline, it has none of those things. It pulls partial data from one system, applies rules from another, and records the outcome somewhere a third system will never read. The result isn't automation, it's the same fragmentation running faster, producing more errors at greater scale.
AI agents operating on fragmented foundations don't automate - they accelerate chaos
When an AI agent touches a fragmented frontline, it doesn't fix the disorder underneath it. It moves faster through that disorder. The agent pulls customer data from one system, applies rules from another, and writes its output back to a third. None of those systems share a common picture of the customer or enforce the same decision logic. The agent acts on partial information - quickly, at scale, and with confidence it hasn't earned.
This is the mechanism behind the pilot-to-production failure. Banks run a contained proof of concept where the context is narrow and the data is clean. Results look good, then they deploy into the real frontline environment where roughly half of the work already lives in the whitespace between disconnected systems. The agent doesn't stall, it keeps moving, but it now follows inconsistent rules and produces decisions that different systems read differently. Every agent a bank adds to a fragmented stack multiplies the coordination overhead - the bank pays more to stay broken at greater speed.
For an AI agent to act reliably, it needs complete customer context, a shared source of truth, and authorized decision authority. On a fragmented foundation none of those conditions exist. That means the AI challenge isn't a talent problem or a tooling problem, it's a structural one. Dropping better models into a broken coordination environment doesn't change the outcome, it amplifies it.
The real prerequisite AI needs that banks aren't building
Most banks treat AI deployment as a tooling problem - pick the right model, hire the right team, run the pilot - but that approach misses the blocker. An AI agent needs complete customer context, a shared source of truth, and authorized decision authority to act reliably. On a fragmented frontline none of those conditions exist. That's a structural problem, and no amount of talent or tooling solves a structural problem.
Without complete context, an AI agent works from a partial picture. It sees the CRM record but not the service history, reads the transaction data but not the open complaint, and issues a recommendation that contradicts what a human told the customer two days ago. The agent isn't broken, the environment it's operating in is. Fragmentation means the AI is always acting on incomplete information, and acting faster than any human would have done it wrong.
The missing infrastructure is a coordination layer that sits above the systems of record - not a replacement for the core or the CRM, but something that makes everything above the ledger work as one. This gives AI a governed, consistent surface to operate on. Across more than 120 bank implementations we've worked through over two decades, that governed surface is the prerequisite that separates deployments that hold at scale from pilots that collapse under production conditions. The standard a bank needs to meet is the one Jennifer Schlossberg, Head of Digital Banking at Synovus, put to the BankingReinvented podcast: "Understand your customer, build a trusted relationship, deliver on your brand promise, and execute with operational excellence every day." AI can only meet that standard when it operates on a foundation designed to support it.
Why a coordination layer above systems of record changes the equation
The structural answer isn't another point solution added to a fragmented stack, it's a layer that sits above cores and CRMs without replacing them. That distinction matters. Cores handle ledger transactions, CRMs track customer history, but neither was built to coordinate work across the frontline in real time. When AI agents need governed authority to act, they need something that speaks to all those systems at once, not a patchwork of API calls across disconnected platforms.
This is what Banking OS does. Banking OS sits above cores and CRMs, routing data and decision rules through a single governed layer so every agent sees the same customer picture. AI agents working inside that layer see consistent data, follow consistent rules, and write back to a single source of truth. Without it, agents operate on partial context and produce inconsistent outputs - faster than before, but wrong in the same structural ways.
Backbase is the only vendor positioned to make this argument concretely. Most competitors address fragmentation by adding integrations, which fixes connectivity but not coordination. A coordination layer governs which agent acts, on what data, under which rules, across which channels - simultaneously. That's a different infrastructure problem, and it's the one that determines whether AI performs at scale across your frontline or just performs well in a controlled pilot.
The compounding cost of staying fragmented as you scale AI
Every AI agent a bank deploys onto a fragmented frontline adds coordination overhead rather than removing it. Each agent needs context, and on a fragmented frontline that context is split across disconnected systems. Each agent either acts on partial data or waits for a human to bridge the gap. Add ten agents and you multiply that problem by ten.
The competitive question for banks has shifted. It's no longer about how the app looks, it's about how the frontline business runs and scales. Banks that stay fragmented will pay a compounding tax with every agent they add. That overhead doesn't stay flat, it grows with the deployment. That's the structural trap most AI strategies are walking into right now in 2026.
Banks will keep doing banking, but the tools supporting that work are changing fast. The banks that move forward are the ones that give those tools a unified environment to operate in - a governed foundation with consistent context and a single source of truth. Without that, more AI just means more speed applied to an already broken coordination model.
Every agent a bank deploys without that coordination layer doesn't reduce the overhead of a fragmented frontline, it adds to it. That bill comes due at scale. Structural clarity is what separates durable frontline automation from accelerated chaos.
Frequently asked questions
What are the biggest challenges banks face when scaling AI beyond pilots?
The core blocker is structural fragmentation, not talent or tooling. Roughly half of frontline work already lives in the whitespace between disconnected systems. When AI agents enter that environment, they operate on partial data, follow inconsistent rules, and write outputs to systems that never communicate. Pilots succeed because context is controlled, and production fails because it isn't.
Why do AI agents in banking produce inconsistent results even when the underlying models are strong?
Strong models still fail when the environment they operate in is fragmented. An agent might pull customer data from one system, apply rules from another, and record results in a third that no other system reads. The model is not broken, the coordination layer beneath it is missing, so the agent acts confidently on incomplete information.
What does a coordination layer in banking do that existing core systems don't?
Cores handle ledger transactions and CRMs track customer history, but neither was built to govern real-time work across the frontline. A coordination layer sits above both, giving AI agents consistent data, shared decision rules, and a single source of truth to write back to. That governed surface is what allows AI to act reliably at scale.
How is the AI challenge in banking different from what it was five years ago?
Banks have used AI in risk modeling for years, so literacy is not the issue. The shift is that banks are now trying to move AI into frontline execution, where the operating environment is structurally fragmented. In 2026, the question is no longer whether to deploy AI but whether the underlying coordination infrastructure can support it without accelerating existing disorder.
What infrastructure does a bank need before deploying AI agents on the frontline?
Three conditions must exist before AI agents can act reliably: complete customer context, a shared source of truth, and governed decision authority. None of those conditions exist on a fragmented frontline. Banks need a coordination layer above their systems of record that unifies those conditions, not another point solution that adds connectivity without solving coordination.
