Agentic banking doesn’t require a core replacement: What 164 corporate treasurers told us about their banks

Over the course of a multi-month strategy and architecture engagement with a major EMEA bank, we surveyed 164 corporate banking professionals across five global markets, conducted 17 in-depth interviews and validated findings with heads of payment and digital banking at Tier 1 banks across Europe and Asia-Pacific.

One pattern emerged from every conversation: Corporate treasurers are under constant pressure to move capital quickly, and the systems they rely on to do so often work against them.

This piece is about that gap — why it exists, why it’s wider in North America than anywhere else and how banks can close it without overhauling their core infrastructure.

The answer is an orchestration layer that sits above legacy systems and gives AI agents a structured, permissioned way to operate within them. It’s not a core replacement. It’s a bridge.

A fish market in a digital world: The current state of North American corporate banking

This experience isn’t an outlier. Every one of the corporate treasurers we interviewed described a version of the same morning.

The technical term for what they’re experiencing is the speed vs. certainty paradox: Treasury teams face constant pressure to move capital and manage liquidity across institutions, but the systems they rely on can’t confirm, in real time, where their money is or whether a payment will clear. The harder they push for speed, the more uncertainty compounds.

Solving this paradox is what AI in corporate banking is increasingly organized around. Understanding where this infrastructure gap comes from is the starting point for knowing how to close it.

The user experience: Speed versus certainty

In markets with harmonized open banking regulation — the revised Payment Services Directive (PSD2) across Europe, comparable frameworks across Asia-Pacific — the paradox largely doesn’t exist. API connectivity gives treasurers real-time visibility into positions across accounts and institutions. Finance teams can move faster because they can see clearly.

But North American corporate banking doesn’t have that infrastructure at the same standard. The market is fragmented, regulation is bilateral rather than standardized and treasurers fill the gap with workarounds that are expensive, insecure or both.

Real-life examples of the insecure methods treasurers use to aggregate data across the 11+ accounts the average professional manages include:

• Screen scraping: Third-party treasury apps use actual corporate login credentials to deploy bots that pull transaction data off bank portal screens. This is a major security vulnerability, and notoriously fragile — a portal UI change can break the integration overnight.
• Manual shadow work: 47% of surveyed users admit to exporting raw data to Excel simply to figure out their daily cash positions. That means nearly half of the world’s most sophisticated finance professionals are doing manual data entry to confirm basic balances. To alert colleagues to payment status, another 47% rely on manual email and 24% use WhatsApp or Teams.
• Legacy batch file transfers: Many organizations still rely on host-to-host (H2H) connections — scheduled end-of-day file drops over secure file transfer protocol (SFTP). Treasurers are completely blind to intraday liquidity until the batch runs.

The result: 76% of users log into two or more banking platforms daily — 64% do so multiple times. 29% of users manage four or more platforms. Nearly half (46%) operate across 11 or more individual bank accounts, with 9% managing over 200.

Nearly half of those surveyed described their platforms using words like “antiquated,” “dated” and “like MS-DOS.” The average net promoter score (NPS) for corporate banking platforms is -12.

The frustration doesn’t end at login. Once a payment is submitted, it enters what treasury professionals describe as a black hole. There is no real-time tracking between “initiated” and “received.” Errors surface days later when a supplier calls to say they haven't been paid.

The commercial outcome: Idle capital, frustrated customers

The financial impact runs deeper than user frustration. When treasurers can’t confirm cash positions in real time, they hold conservative liquidity buffers across accounts as insurance against delays they can’t see. That’s idle capital that could be deployed or earning yield. When clients can’t resolve inquiries independently — because the platform gives them no visibility into payment status — they call the bank. Every one of those calls is operating expense (OPEX) that a well-built platform eliminates.

The autonomous future of corporate banking: From assisted to agentic

What agentic banking actually means: Agentic banking is corporate banking infrastructure where AI acts on behalf of the client rather than waiting for instructions.

Instead of a treasurer logging in to check balances, flag anomalies and initiate transfers manually, an AI agent monitors their liquidity position continuously, anticipates gaps and either surfaces recommendations or executes routine transactions autonomously — within defined, permissioned parameters.

This is different from what most banks currently describe as AI, and the distinction matters.

The current model: Assisted workflows

Today’s best banking platforms offer assisted workflows. They flag a data entry error before submission. They suggest a payment template. They surface a dashboard with consolidated positions. These are useful, but they are not agentic.

Assisted workflows still require a human to monitor, recognize a problem and act. The bank remains a passive utility — available when asked, reactive by design. A treasurer still starts their morning by extracting balances from 35 portals. The tool just makes some of that extraction slightly less painful.

The future model: Agentic AI

Agentic banking changes the bank’s role. An AI agent with continuous access to real-time transaction data can predict an upcoming liquidity shortfall and surface the right working capital facility before the treasurer realizes a gap is forming. An AI agent with read access to a client’s enterprise resource planning (ERP) software can reconcile invoices, auto-stage multi-currency payment runs and flag compliance issues — without manual initiation.

Research across surveyed users identifies five use cases with the clearest ROI:

• Data-driven cross-selling: Deploying live transaction monitoring to trigger timing-sensitive lending, trade and FX solutions at the exact point of customer need.
• Agentic ERP integration: Applying Model Context Protocol (MCP) to bi-directionally read ERPs, auto-stage payments and manage tokenized assets.
• Reduced cost-to-serve: Implementing unified visibility and real-time transaction tracking removes the visibility “black hole,” significantly decreasing the volume of manual client inquiries and associated OPEX.
• AI-assisted risk mitigation: Deploying live anomaly identification and automated verification before submission to safeguard liquidity and drive higher straight-through processing volume.
• Omnichannel delivery: Creating a consolidated digital entry point across global markets, providing full-scale desktop functionality alongside mobile speed secured by biometric authorization.

Each of these maps directly to documented pain points. Automated reconciliation ranked as the single most valuable automation feature across all users surveyed. Anomaly detection before submission ranked in the top three for 53% of those surveyed.

The demand is confirmed. The infrastructure to fulfill it is what’s missing.

The commercial case for agentic banking

For the bank, the shift from passive utility to active partner means a larger share of wallet earned through timing and relevance rather than generic product promotion. A foreign exchange (FX) hedge recommended during a large pending cross-border payment isn’t a sales push — it’s a service the treasurer needed anyway. That difference defines the relationship.

On tokenization

The same orchestration architecture that supports agentic workflows today is built to accommodate tokenized assets as they come online. Digital twins of commercial deposits, trade instruments and bond holdings can flow through the same orchestration layer without a core rebuild.

Banks that build the bridge now don’t re-architect when settlement moves to 24/7 ledgers. Instead, they extend what’s already in place.

The roadmap to agentic banking: Build the strategic bridge

The challenge for banking CTOs is that you can’t run an agentic AI agent directly on a 40-year-old mainframe. The core keeps the lights on. It runs settlement, processes transactions and stores the records that regulators audit.

Replacing that core is a multi-year project that ties up IT budget, delays every client-facing feature and introduces risk most institutions can’t absorb.

So, what’s the alternative? Build above it.

The technical bridge: Model Context Protocol (MCP)

A functional backend-for-frontend (BFF) layer serves as an orchestration tier between your legacy core and modern AI tooling. By implementing standards such as the Model Context Protocol (MCP), this layer gives AI agents a structured, permissioned way to read ERP data, reconcile invoices and pre-stage payment runs — all without touching the underlying infrastructure.

The mainframe keeps doing what it does. The orchestration layer handles what it can’t: real-time data aggregation, AI-readable context, bidirectional ERP connectivity and the presentation logic that lets client-facing teams deploy new features without waiting on a core modernization backlog.

New capabilities ship in weeks rather than queuing behind a transformation program. IT spend shifts from maintaining existing systems to building the next layer of capability on top of them.

The trust bridge: Security for an agentic world

Agentic workflows don’t remove humans from decisions — they remove humans from the wrong decisions. High-value, edge-case transactions require human judgment. Routine, rules-based operations don’t.

Security architecture reflects this. Mobile biometrics handle executive authorizations. An approval that once required a physical card reader now takes seconds from a phone. AI anomaly detection surfaces potential fraud, compliance issues or data entry errors before a payment is submitted. Straight-through processing rates go up while call-center volume goes down. The speed versus certainty paradox doesn’t just improve, it resolves.

How TELUS Digital helps banks build the bridge

Most vendors approaching this problem lead with a platform — a product that requires displacing what you have. TELUS Digital’s position is different. We build the integration layer, not the replacement.

Our work sits between your legacy core and the AI tooling you want to deploy, connecting them without touching your settlement infrastructure or compliance architecture. The goal at each engagement is the same. You get working agentic capability as quickly as possible, starting with the use cases with the clearest ROI.

Starting with a focused proof of concept — before a program is scoped — is consistently how the most defensible cases for larger investment are built. Our two-week Agentic AI Accelerator targets the friction point with the most quantifiable ROI in your environment and puts a working MCP-based capability in front of your team before any larger scope is on the table.

See what the Accelerator program covers.