BNY Eliza Platform: Enhancing AI Solutions with OpenAI

BNY's Eliza platform uses OpenAI technology to help employees build controlled AI agents for specific tasks. With over 20,000 staff enabled to create and use agents, BNY aims to automate routine work, enhance consistency, and allow teams to focus on higher-value client outcomes—while maintaining robust controls.

Enterprise AI has moved beyond pilot stages. The real differentiator now is whether an organization can safely scale adoption—across teams, geographies, and risk profiles—without creating chaos.

BNY's approach provides valuable insight for any large enterprise: create a platform (Eliza), integrate governance, and empower people closest to the work to develop AI agents that address real problems. According to OpenAI’s case study and BNY’s own resources, Eliza is designed as an enterprise AI platform to enhance client service, operations, and cultural transformation—while supporting an “AI for everyone” model of adoption.

What is the Eliza platform at BNY?

Eliza is BNY's enterprise AI platform, created to offer reusable AI capabilities across the organization. BNY describes it as being designed to improve client service, boost company operations, and drive cultural transformation using AI.

What makes Eliza unique isn't just the tools—but the combination of:

• Widespread enablement (a large number of employees can build and use agents), and

• Guardrails (identity, access controls, workflow boundaries, and governance suitable for regulated work environments).

What's new: 'AI agents' and BNY's 'digital employees'

In OpenAI's article, BNY describes advanced agents as 'digital employees'—AI agents with identities, access controls, and dedicated workflows. These agents can be tailored to specific processes, moving humans from performing every first draft of work to supervising, training, and enhancing the agent’s output over time.

This perspective is important because it alters how you plan adoption:

• You're not just deploying a chatbot.

• You're introducing agentic workflows that interact with data, decisions, and operational steps—so governance needs to be designed in, not added later.

Why this matters for efficiency (and client outcomes)

Efficiency gains don't come from 'using AI'. They arise from redesigning workflows so that:

1. routine tasks are automated,

2. quality checks are consistent, and

3. people spend more time on judgment, exceptions, and client-facing value.

OpenAI’s case study cites examples of agents supporting tasks such as payment instruction validation and code security improvements. These are precisely the types of tasks where automation can reduce friction while enhancing consistency—two aspects that clients notice quickly (fewer delays, fewer avoidable errors, faster turnaround).

There's also evidence that BNY has promoted broad AI literacy: Fortune reported that a BNY spokesperson mentioned 98% of employees were trained on generative AI, with many using Eliza daily (as of September 2025).

How Eliza supports enterprise-wide adoption

Most AI programs stall because adoption is treated as a communications initiative rather than an operational model. Eliza's strategy—enablement plus guardrails—aligns with what truly works in large organizations:

1) Make building 'allowed' (and supported)

If only a central team can build, the backlog grows uncontrollably. By enabling a large number of employees (reported as 20,000+) to build agents, BNY greatly expands capability—while keeping development aligned with real operational needs.

2) Standardize the platform layer

A unified platform avoids numerous disconnected tools, prompts, and shadow workflows. BNY positions Eliza as a foundational set of reusable capabilities across the enterprise.

3) Add identity and access controls for agents

Once agents become 'digital employees', you need clear boundaries: what they can access, what they can do, and how activity is monitored. OpenAI highlights identities and access controls as crucial to BNY's concept.

4) Use 'deep research' and structured reasoning where applicable

OpenAI notes that certain teams are experimenting with ChatGPT Enterprise capabilities like deep research for multi-step reasoning across internal and external data—useful for scenarios like strategic planning and analysis.

(Practical takeaway: don’t force agents into every task. Use them where multi-step work or repetitive processing offers meaningful leverage.)

Practical examples (the kind that typically work first)

If you're trying to replicate this in your own organization, start with tasks that are:

• repetitive and rules-based,

• high-volume,

• low-to-medium risk with robust review loops, and

• troublesome enough that teams actually desire change.

Common starting points we see in enterprise workflow tools include:

• Client reporting drafts and summaries (human-reviewed before sending)

• Request intake and triage (categorize, route, assign, set SLAs)

• Knowledge Q&A (anchored in internal policies and playbooks)

• Document checking (flag missing fields, inconsistencies, exceptions)

In regulated environments, the pattern that proves effective is “AI does the first pass; humans handle exceptions and approval”.

What other enterprises can learn from BNY’s playbook

BNY's narrative is compelling because it illustrates three principles that apply broadly:

1. Platform thinking beats point solutions
   A single enterprise AI platform creates consistency, shared controls, and reusable building blocks.

2. Adoption scales when you treat employees as builders
   Empower those familiar with the work to shape agents—paired with training and governance.

3. Governance is the feature, not the afterthought
   Identity, access, auditability, and policy gain importance as you move from “chat” to “do”.

How Generation Digital helps you apply this (without the exaggeration)

If you appreciate the pattern (platform + governance + adoption) and want to apply it to the tools your teams already use, Generation Digital can help you design and implement AI-enabled workflows safely and effectively—especially across Asana, Miro, Notion, and Glean.

Where we typically start:

• AI readiness and roadmap: prioritize use cases that will succeed, define owners, and establish success measures.

• Workflow design and automation: eliminate operational bottlenecks with clear transitions and automation patterns (intake → triage → delivery).

• Trust and governance: align data residency, security controls, and admin policies so adoption doesn't stall at