How a 700-Person Bank Graduated From Simple AI Chat to Always-On AI Agents

Talk about miracles: Joining an internal committee at a 700-person bank became the most sought-after assignment within the company.

Here’s why:

Peapack Private Bank & Trust was bringing together employees from across its departments to identify problems AI could help solve. Chief Technology Officer John Kowal says the AI champions group actually became competitive to join.

“People wanted to join, they wanted to make AI a part of their careers even early on.”

For a private bank with about 700 employees, that interest mattered. Peapack serves individuals, families, and businesses in the New York tri-state area through personal banking relationships. Its people knew where repetitive work was getting in the way of that mission. The champions program gave them a way to do something about it.

And boy did they do something about it.

Those employees, guided by leadership, helped move the bank from having the typical broad access to AI chat that you typically see at companies to something else entirely.

Today, the company uses AI to build their own custom software and workflows, and has even rolled out AI agents that it calls “digital employees,” or systems that autonomously handle recurring work and contact people when something needs attention.

In Episode 238 of The Artificial Intelligence Show, part of AI Transformations presented by Google Cloud, Kowal explains how the bank made that progression.

And it all began way back in 2023…

First, Give Everyone a Way to Learn

Peapack began exploring generative AI while the technology was still new to most businesses. Its initial conclusion was that the potential was large enough to warrant learning together.

The bank put policies and governance in place, trained employees, and provided tools for AI chat and meeting summaries. By the end of 2023, every employee had access to those capabilities, Kowal says.

That gave people a starting point in work they already recognized, such as researching a question, writing, or catching up on a meeting. It also gave the bank experience with what the technology could actually handle.

Some early attempts disappointed. Connecting AI to documentation and expecting helpful answers did not work well for Peapack in 2023. The bank tried again as the technology improved and its team learned more. That type of persistence paid off: answering questions from documents is now a core part of its regular AI usage, according to Kowal.

Let the Departments Bring the Problems

Importantly, CEO Doug Kennedy backed the transformation from the beginning and proposed the AI champions concept. The bank built a network of 29 people embedded in its different business divisions.

Those champions changed where the work originated. The technology team had initially introduced tools and possibilities. Now, the people doing the banking, marketing, support, and other departmental work could bring forward the problems they wanted solved.

Kowal describes the invitation this way:

“You don't need to bring us the answer, but bring us the challenge that your department's having and let's solve it and see how we can use AI to solve it.”

The distinction made participation possible for employees who understood a problem better than they understood the technology. It also left room for an answer involving conventional automation when that fit the work.

The bank supported that flow of ideas through monthly individual meetings with champions about their projects. In quarterly technology innovation webinars, champions showed colleagues what they had built and what had changed.

Those demonstrations helped other teams apply what a department had learned. Employees could recognize a process from another department and see where the same approach might help their own team. Over time, some champions also began building simpler tools, such as chatbots, themselves.

At the program's first anniversary, Kowal reports that the champions had completed around 100 projects, with more than 80 additional projects in the pipeline.

Give the Ideas Something to Build On

As the bank's ambitions grew, its earlier technology investments became more consequential.

Peapack had spent years previous to the release of ChatGPT in late 2022 developing a data warehouse, or a shared store of business information. It also had employees with software development experience. Those resources gave its AI work access to both company data and people who could turn an idea into working software.

So, as the generative AI revolution picked up steam, the bank established a dedicated AI engineering team. Its developers began using Codex, OpenAI’s agentic coding tool, to build applications tailored to the business, including a platform for the wealth division's account reviews.

The data team gained a different capability. A system called Project Atlas uses knowledge of the warehouse's structure and the bank's operations to turn questions into code that can query the data. Kowal says analysts can now return answers during meetings with executives, changing how quickly a business question can become a useful answer.

These were different kinds of progress. AI could help someone develop an application. It could also help a specialist work with existing data. Both gave the bank more ways to act on the problems its departments were surfacing.

The champions supplied the business context. The data and development teams made more of their ideas feasible.

Move From Individual Requests to Recurring Work

The next shift concerned how employees interacted with AI.

Chat tools required a person to initiate a task. Peapack also wanted systems that could carry out an ongoing responsibility, watch for an issue, and bring it to someone's attention.

The bank began combining AI agents, traditional automation, internal data, and existing tools into what it calls its “digital employees.” Some of these digital employees even have their own computers to run the tools their assignments require.

Oh, and they all have names.

In IT, “Alex” monitors the bank's ServiceNow support tickets. Alex offers an initial suggestion, routes requests to the appropriate specialists, and keeps watching the queue for issues that need escalation.

In marketing, “Mia” reviews the bank's websites each week for outdated content and problem links. She also suggests ways to make the sites easier to find.

These systems communicate through Microsoft Teams and Outlook. An urgent finding can arrive as a Teams message; routine information can arrive by email. Employees can then reply and continue the conversation.

That changes the employee's role in starting the work. Someone no longer has to remember to ask for every website check or every pass through the support queue. The recurring assignment gives the system something to do between conversations.

In this way, Peapack also keeps its existing safeguards in place. AI can double-check work and flag concerns, but it cannot substitute for the checks the bank requires.

Measure What Changes for the People Doing the Work

As more tools reached employees, the bank needed to know whether they were improving the work.

For “Penny,” a digital employee that answers questions from the bank's financial centers, counting chatbot conversations would have been an easy place to start. Peapack looked instead at demand on the team that supports those centers.

Kowal reports that the support team's tickets fell 60% over the four months since Penny launched.

His larger test is simple and direct:

“Are we saving time? Fine, but are we actually producing more with that time?”

The question keeps the bank focused on the purpose of the work. A useful system should change what employees can accomplish, how quickly they can respond, or how much attention they can give a client.

The Playbook Is a Sequence of Increasing Responsibility

There’s a lot to learn from how John has ushered Peapack through this multi-year AI transformation. And the bank’s progress suggests a practical sequence for leaders moving beyond individual AI use:

1. Start with supported access and real learning. Give people useful tools, training, and clear boundaries.
2. Bring departmental problems into the process. Let employees contribute their knowledge of the work before asking them for technical answers.
3. Keep examples and support recurring. Make time to discuss projects and show other teams what is working.
4. Build the capacity to act. Develop the data access and technical skills that more ambitious work requires.
5. Expand from requests to responsibilities. Identify recurring work that a system can perform and situations it should bring to a person.
6. Judge the result through changed work. Measure the outcome for the team or customer the system is supposed to help.

Peapack is still expanding its digital employees across the bank. The goal is to give them more knowledge, access to more platforms, and useful responsibilities alongside more employees.

The enthusiasm for joining the champions program was an early human signal. The bank made that enthusiasm useful by giving employees a route from noticing a problem to changing the way their department worked.

Did you enjoy this transformation story? Go deeper on how AI is reshaping work and business with The Artificial Intelligence Show. Each week, we break down what matters in AI and what leaders should do about it.