I recently reported on a major shift inside one of the world’s largest financial institutions. BNY has embraced AI across its sales organization and is now running with thousands of advisors who are AI enabled. The results are measurable: faster, smarter account planning; richer client conversations; and a tangible increase in time spent on high-value advisory work. This is not theoretical change. It is a practical transformation in how teams prepare, collaborate, and deliver for clients.
📣 The big picture
BNY’s sales organization brought AI into everyday workflows to help relationship managers and advisors do more of what humans do best: understand clients, craft tailored strategies, and nurture long-term relationships. The initiative is enterprise scale — more than 50,000 employees are AI enabled — and aims to redesign the routine work that steals time from client-facing activity.
The central idea is simple. Instead of spending hours assembling background information, pulling disparate data, drafting account plans, and hunting for the latest research, advisors get an AI-powered starting point. That starting point is a living, up-to-date account template with the profile, context, and research already integrated. Instead of replacing human judgment, this approach augments it. Advisors can spend less time on administrative preparation and more time on strategy, insight, and connection.
🔍 What changed in the day-to-day
I heard a clear refrain from the team driving the project: efficiency has freed up attention. One leader put it bluntly:
“All 50 plus thousand employees are all AI enabled.”
That achievement is not just a PR line. It reveals a purposeful roll-out that touches multiple roles and responsibilities. The visible changes in daily work include:
- Faster account planning — The time to assemble a client plan has fallen dramatically.
- Richer client profiles — AI consolidates internal data, public filings, market news, and prior interactions into a single view.
- Living documents — Account plans are no longer static PDFs or slide decks. They are dynamic artifacts that update as conditions change.
- Focused client time — With basic research automated, advisors spend more minutes and more mindshare on strategy and relationship building.
📉 Numbers that matter
Concrete metrics make the transformation tangible. The team reports a sweeping improvement in operational speed. One of the most striking figures I recorded was their reduction in planning time:
“We are seeing a 60% decrease in the amount of time it takes to put that plan together.”
A 60 percent reduction is remarkable in a field where planning typically requires cross-checking multiple systems, asking for internal memos, and manually reconciling client notes. That level of efficiency translates directly into time that advisors can use to prepare for strategic conversations, develop new solutions, or simply deepen client relationships.
🤝 From tools to relationships: client impact
Speed matters, but so does quality. The shift is not about rushing through tasks. It’s about giving advisors better context more quickly so they can ask higher-quality questions and offer more relevant solutions. As one leader framed it, when you can run the company better behind the scenes, you can come up with innovative solutions for clients.
Several aspects of client impact stood out:
- More grounded conversations — Advisors arrive prepared with up-to-date analysis and tailored talking points.
- Timely responses — With live research and alerts, teams can respond to market events or client changes faster.
- Personalized strategies — AI synthesizes signals that lead to more relevant recommendations for each client.
- Higher touch where it counts — AI takes on the routine so human effort goes toward strategy and care.
That last point matters more than numbers. Investment and treasury advice is fundamentally a relationship business. AI’s most valuable role here is enabling more meaningful attention.
🧠 The mechanics behind the magic
People often want to skip from benefit to conclusion without understanding the mechanics that enable it. I dug into how the capability is built and applied.
At the core are a few repeatable technical and process components:
- Data consolidation — The system ingests internal CRM records, account histories, public filings, news feeds, and proprietary research. That consolidated view forms the single source of truth for an account.
- Contextual prompts and templates — Advisors start with smart templates that pre-populate sections of an account plan based on the consolidated data. The templates are configured for different client types, business lines, and advisory services.
- Retrieval-augmented generation — Instead of relying on static model outputs, the system retrieves the latest documents and data to ground its responses, reducing hallucination and improving relevance.
- Agentic models — The platform connects language models with tools and workflows. These agentic models can run searches, pull documents, summarize findings, and even suggest next steps for an advisor to review and approve.
- Living document architecture — Account plans live in a system that allows ongoing updates. When new market developments occur, profiles refresh and alerts flag advisors to changes that matter.
- User feedback loop — Advisors continually correct and refine outputs. That human-in-the-loop feedback improves future outputs and helps the system learn prioritization and tone preferences.
Taken together, these components turn manual, one-off research sprints into a continuous, automated intelligence layer that supports human advisors.
🧩 What “Eliza” does
One name that came up in conversation encapsulates the service layer: Eliza. In their implementation, Eliza automates a cluster of repetitive planning tasks. The description was practical:
“Now Eliza does that for you. So imagine handing a template of an account plan over and that profile and the most up-to-date information is already there.”
That sentence captures the user experience: a template handed off, prefilled with the richest available context. Eliza is not a black-box agent taking decisions out of human hands. It is more like a highly capable research assistant that delivers drafts, insights, and prioritized next steps for human review.
Because the output is a living document, the draft is not static. It evolves as the system ingests new intelligence, and advisors can pull or push updates as needed. The combination frees advisors from repetitive assembly work while keeping them fully in control of strategy and client-facing narrative.
🧭 Agentic models: what they are and why they matter
The team described their models as “agentic,” a term that can sound academic but is straightforward in practice. Agentic models are systems that can take multi-step actions on behalf of a user, orchestrating tools and data sources to achieve a goal. They don’t replace human oversight; they orchestrate workflows. For example:
- Search multiple databases for the latest filings and news related to a client.
- Summarize the findings into a two-page briefing with highlights and risk signals.
- Update the account plan, flag sections for advisor review, and schedule a follow-up reminder if a material change is detected.
Putting this into practice requires disciplined design. The models must be connected to the right tools, given guardrails for safe action, and designed so that final decisions remain human-approved. The team emphasized that partnership and advisory expertise were essential when building these agentic behaviors:
“This relationship allows us to know that we are not only being able to leverage the latest but also getting great advice on how we're building these agentic models.”
That line highlights two truths: that the technology is evolving quickly and that successful deployment requires external technical and governance counsel.
⚖️ Governance, compliance, and trust
Financial services operate in a highly regulated environment. Any AI deployment must address confidentiality, data residency, auditability, and compliance. BNY’s approach included several safeguards:
- Data access controls — Role-based permissions ensure that advisors only see the data they are authorized to access.
- Auditable actions — Agent steps, summaries, and edits are logged so supervisors can trace how a recommendation was produced.
- Human-in-the-loop review — AI drafts are explicitly presented as advisor-ready materials that require sign-off before client delivery.
- Vendor partnership — Close collaboration with technology partners provided both technical support and compliance guidance.
Trust is also cultural. Teams must believe that AI will help them do their jobs better rather than replace them. The governance model needs to protect clients and the bank while preserving advisor autonomy and accountability.
🛠️ Implementation playbook: how they rolled this out
Turning an idea into enterprise practice requires an execution playbook. The rollout at BNY combined technology, people, and process changes. Here is a distilled playbook that others can adapt:
- Identify high-impact workflows — Start where the time savings are obvious: account planning, portfolio review briefs, or recurring client reporting.
- Assemble a small cross-functional team — Include compliance, IT, advisors, and product owners. Early advisory input keeps the product usable and trustworthy.
- Build a pilot with real users — Start with a small group of advisors who volunteer. Rapid cycles of feedback are essential.
- Connect authoritative data sources — The pilot fails if the AI is disconnected from the right data. Ensure direct access to enterprise systems and trusted external feeds.
- Create templates and guardrails — Pre-built templates accelerate adoption and ensure consistency in tone and scope.
- Design for explainability — Provide visibility into why the AI suggested a particular recommendation or flagged a risk.
- Measure outcomes — Track time savings, client satisfaction, and advisor adoption. Use those metrics to iterate and expand.
- Scale thoughtfully — Gradually expand to more users and more workflows, adjusting governance and training along the way.
This playbook is intentionally pragmatic. The emphasis is on reducing friction and ensuring that advisors trust what the system delivers.
🧑🏫 Training and adoption
Adoption rarely happens without investment in people. The organization paired the technical launch with robust training and change management. Practical tactics included:
- Hands-on workshops — Advisors used the tool in simulated client scenarios to learn its strengths and limitations.
- Office hours — Product and compliance owners held open sessions to answer questions.
- Playbooks and cheat sheets — Quick guides explained when to rely on AI drafts and when to double-check specific sections.
- Champion networks — Early adopters became internal champions who supported their peers.
Training focused on one core message: AI is an assistant, not an autopilot. Advisors who internalized that principle were able to leverage the tool's speed while preserving the judgment that clients trust.
🔄 The ongoing feedback loop
Building in a feedback loop matters as much as the initial development. Advisors continuously corrected and refined outputs, which fed back into model tuning and template updates. That loop looks like:
- Advisor reviews AI draft and leaves edits or comments.
- Product team aggregates common edits and updates templates or prompts.
- Model updates and guardrails are adjusted to reflect real-world usage.
- New output quality is monitored and shared with end users.
This cycle is how a system goes from "helpful but rough" to "trusted partner." It also ensures that the system evolves in ways that reflect real work practices rather than hypothetical requirements.
🔬 Evidence of improved client experiences
The proof is in client outcomes. Advisors report that the combination of speed and relevance leads to better client experiences in several ways:
- Faster response windows — Clients receive insights and proposals more quickly after market changes.
- More tailored proposals — Recommendations reflect client history and current context rather than generic options.
- Higher-quality check-ins — Advisors arrive at meetings with a focused agenda and deeper understanding.
- Stronger relationships — Time saved on preparation is reinvested in relationship building and proactive strategy.
Advisors consistently noted that clients appreciate having an advisor who is informed, proactive, and ready with actionable ideas — not just reports. AI helps deliver that experience repeatedly and at scale.
🧾 A sample workflow: from alert to client action
To make this concrete, here is a simplified workflow that illustrates the value chain from market event to client action:
- Trigger — A macro event or company filing matches an advisor’s watchlist.
- Automated research — The agentic model aggregates the latest documents, news, and portfolio impact analyses.
- Draft brief — A two-page brief is prepared that includes implications and proposed talking points.
- Advisor review — The advisor edits the brief, adds personal context, and approves next steps.
- Client outreach — The advisor contacts the client with a timely, personalized recommendation or question.
- Follow-up — If a client chooses a recommended action, the system can route the request to operations or create a task for execution.
In this workflow, the AI handles the heavy lifting of assembly and initial analysis. The advisor adds judgment and human context, then quickly converts insight into action. The process shortens reaction time and raises the quality of the recommendation.
🧾 What this means for sales strategy
From a sales strategy perspective, the implications are strategic and operational. Advisors with access to rapid, well-organized intelligence can pivot from product pitching to solution design. This enables:
- Deeper relationship selling — Conversations become consultative and centered on client outcomes instead of product checklists.
- Cross-functional collaboration — With a shared account plan, product specialists and service teams can act in concert.
- Scalable personalization — Tailored advice becomes feasible across large client populations.
For sales leaders, the tactical choice is to amplify what humans do well — counsel, negotiation, empathy — while automating the background tasks that slow them down.
🧱 Technical considerations for builders
If you are building something similar, there are a few technical considerations to keep front and center:
- Data freshness — Make sure the retrieval layer can fetch the latest documents and feeds quickly.
- Model grounding — Use retrieval-augmented generation and explicit source citations to reduce hallucinations.
- Tool integration — Connect scheduling, CRM, and execution systems so AI suggestions can become actions.
- Security — Encrypt sensitive data and enforce strict access controls.
- Scalability — Plan for the concurrency and throughput of agents acting on behalf of tens of thousands of users.
Addressing these areas early reduces surprises and improves adoption speed.
💡 Lessons learned and practical advice
From the deployment I examined, several lessons stood out that others can adopt:
- Start with the advisor, not the model — Design for the human workflow first, then automate the parts that are repetitive and rule-based.
- Be explicit about limits — Ensure every output is presented with context about data sources and confidence levels.
- Keep humans in control — Make final approvals easy and log the decision trail for compliance and quality control.
- Measure what matters — Track time savings, client satisfaction, conversion rates, and error rates to guide investment decisions.
- Iterate fast — Regularly release template updates and gather user feedback to keep the system aligned with needs.
Those practical habits reduce friction and ensure the technology actually helps people do their jobs better.
📈 The long arc: from efficiency to innovation
There is a strategic arc here. The first wave of value comes from efficiency: faster planning and better use of advisor time. The second wave is innovation. Once the organization removes the burden of repetitive tasks, it can redeploy that human capital toward designing new client products, building bespoke solutions, and experimenting with service models that were previously impractical at scale.
One of the leaders summed this up succinctly: when you are able to run your company better, you come up with innovative solutions for our clients. That is the real promise. Efficiency is not the end goal; it is the platform for deeper client-centered innovation.
🔮 Where this leads next
Looking forward, I expect the following developments as teams scale and refine these systems:
- More proactive playbooks — Systems will anticipate client needs and automatically suggest multi-step playbooks for advisors.
- Expanded agentic workflows — Agents will coordinate across internal teams to execute parts of a plan once an advisor signs off.
- Better personalization at scale — Sophisticated segmentation and model tuning will make personalized advice feasible across thousands of accounts.
- Tighter regulatory integrations — Automated compliance checks and explainability features will be built into workflows to reduce audit friction.
These are not speculative features. They are logical extensions of the living-document, agent-enabled model that BNY has begun to operationalize.
🧾 Final thoughts
I saw a practical, human-centered application of AI that respects the norms of financial advice while giving advisors the tools to be more effective. The emphasis on living documents, agentic assistance, and a measurable reduction in planning time shows that AI can be an operational multiplier without eroding the human relationships that financial services are built on.
Change of this kind is never purely technical. It is also cultural, managerial, and procedural. But with a clear playbook, strong governance, and a focus on advisor enablement, I believe organizations can replicate this success. For teams looking to make the shift, the clear advice is to prioritize data consolidation, design for human approval, and measure the outcomes you care about — client trust and advisor time.
In short, the combination of AI and human judgment can make every client interaction more timely, relevant, and valuable. That is the outcome that matters most.
