BNY Frontier case study
What happens when America's oldest bank hands repetitive work to AI? This customer story infographic shows how BNY built an internal AI platform to run "digital employees" inside the same systems, workflows, and audit trails that govern its human workforce. Read the infographic to see how BNY reshaped its organization around AI without sacrificing control.
What are BNY’s “digital employees” and how do they work?
BNY uses the term
“digital employees” to describe AI agents that work inside the same systems, workflows, and audit trails as human staff.
These agents run on BNY’s internal AI platform,
Eliza, which is built partly on Microsoft Azure. Instead of sitting on the sidelines as a separate tool, digital employees:
- Have unique IDs, credentials, supervisors, and reporting lines—just like regular employees.
- Log every action in an audit trail, so their work is fully traceable and reviewable.
- Operate within a secure, credentialed infrastructure with identity management and encryption.
BNY uses these agents for
repetitive, rules-based tasks that run around the clock, such as payment validations inside regulated financial systems. This allows the bank to keep human judgment and accountability at the center, while letting AI handle high-volume execution.
For a bank that clears most US Treasury bills and safeguards
$59.4 trillion in assets for institutions worldwide, this setup helps ensure that AI is
observable, controllable, and auditable before it ever touches live, high-stakes processes.
How is AI changing roles and the org structure at BNY?
BNY is using AI to
reimagine how work is structured, not simply to cut costs.
Historically, the organization looked like a
pyramid: a wide base of people doing repetitive processing work, with fewer employees focused on analysis and strategy. As digital employees take on more of the routine tasks, BNY expects that shape to move toward a
“diamond”:
- The base of repetitive work is handled by AI agents.
- The middle layer of analytical and creative roles expands.
- Strategic leadership at the top continues to make the key decisions.
BNY is explicit that the goal is to
reshape the organization, not shrink it. Even as it adds more agents, the bank still plans to
hire thousands of human employees every year.
As routine processing shifts to digital employees, human roles move:
- From processing to problem-solving.
- From execution to strategy.
- From clearing queues to benchmarking clients and designing new agents.
In other words, the organization becomes more of a
learning system, where intelligence compounds over time and people are freed up to focus on higher-value work.
What measurable impact has BNY seen from AI so far?
BNY tracks the impact of AI not just through ROI, but through
time saved,
errors reduced, and
capacity freed for higher-value work.
In one key area—manual payment validations—the bank reports that:
- Time per validation dropped from 5–6 minutes to under 30 seconds.
- Open investigations fell from about 5,000 to fewer than 1,000.
- Capacity was freed up for analytical work across time zones, rather than just queue-clearing.
These gains are supported by a significant commitment to change: BNY is investing
nearly 19% of its annual revenue in transformation, including secure infrastructure, AI governance, and workforce evolution.
BNY also emphasizes that every AI action is
traceable. Audit trails do double duty: they satisfy regulatory and risk requirements, and they also serve as
learning signals to improve both agents and processes over time.
Overall, the bank measures
momentum—how quickly it can redesign work and move people into more strategic roles—rather than looking only at short-term cost savings.
BNY Frontier case study
published by Mayhem Shield
More about us
Mayhem Shield is an independent, buyer-side assurance practice for enterprise AI deployments. When an organization is preparing to approve an AI tool for production, a coding assistant, a RAG pipeline, an agentic system, its approval forums need evidence of how the implementation will actually operate in that environment, not a vendor marketing pack. That evidence is what we produce.
We do not sell, implement, or operate the AI products we review. We are paid only by the buyer, never by the vendor. That separation is the product: it is what makes our findings defensible in front of security, architecture, risk, and audit stakeholders.
How we work
- Structured, repeatable review logic. Phases, evidence rules, severity calibration, and gate criteria are defined in advance, not invented per engagement. The methodology is published and inspectable on GitHub without a sales call.
- Grounded in your environment. Findings are tested against your identities, data paths, integrations, and workflows as actually deployed, not against the vendor's reference architecture.
- Decision-ready outputs. Every engagement ends in a written position: go, conditional go, or no-go, with a traceable findings register, evidence requests, and conditions tied to POC, pilot, and production gates.
Core capabilities
- AI implementation assurance reviews. Fixed-structure packages from a two-week rapid readiness review of one tool through a portfolio program covering three or more tools under one assurance standard.
- Architecture and trust-boundary analysis. Deployment model, data flow, identity, and integration scope for AI systems, documented in formats governance forums already recognize.
- AI vendor claim verification. Assessment of whether a vendor's published security and data-handling claims are checkable, contractual-only, or unverifiable, before those claims underwrite an approval.
- Security and governance advisory. Buyer-side support for AI review boards, evidence standards, and approval gate design.
We maintain relationships with major cloud and technology providers for market and technical visibility. Because our work is buyer-side assurance, we take no resale margin or implementation fees from any vendor, and any relationship relevant to a specific review is disclosed to the client at scoping.
Our commitment
Approvers carry personal and organizational risk when they sign off on an AI deployment. Our job is to make sure they sign with evidence in hand. For more information, visit www.mayhemshield.com or contact us at info@mayhemshield.com.
Sign in with LinkedIn