How Greylock Federal Credit Union Is Building Toward Agentic Orchestration with Camunda
Download PDF
EARLY SIGNALS
~7%
improvement in efficiency ratio, met nearly a year early
100+
processes orchestrated, the foundation for AI
Rules + AI + people
in one governed, traceable process
AI agents
next on the near-term roadmap
About Greylock Federal Credit Union
Greylock Federal Credit Union is a member-owned community credit union headquartered in Pittsfield, Massachusetts, serving Berkshire County and the surrounding region. It builds and maintains custom applications across its core banking systems, and has been an early mover in bringing process orchestration, and now AI, into the operations its members rely on. Capital BPM, its implementation partner, has worked alongside the team to accelerate the platform foundation this work builds on.
Where They Were, Where They’re Headed
Greylock has brought more than a hundred processes under orchestration with Camunda, and AI agents are now on the roadmap, taking on the judgment-heavy work while a person stays in control of every high-stakes decision.
WHERE THEY WERE
Bringing AI in would have meant handing decisions to a black box the team could not audit or fully trust.
Address-hygiene checks leaned on rigid rules, with manual review whenever something looked unusual.
Automation ran the steps, but people still made every judgment call and correction by hand.
Audit tracking was assembled manually, one process at a time.
WHERE THEY’RE HEADED
The model they are building has a language model score each case while a person validates the result, so speed never costs them control.
Deterministic rules will handle the routine while AI takes the hard cases, and once scoring is trusted, an AI agent will make the corrections.
AI agents will route routine corrections automatically, reserving people for the high-stakes calls.
Consistent audit logging on every step is being built in, so more can be safely handed to AI agents, with a complete, reviewable trail.
Why It Mattered
Everything Greylock automates comes back to its members. As a community credit union, the processes running behind onboarding, servicing, and daily operations shape what members feel, and slow or inconsistent work shows up in their experience. It runs more than a hundred orchestrated processes with hundreds of executions a day, and as a regulated institution, each one has to be documented, consistent, and defensible whenever a regulator asks.
When the business first came with an urgent need, process orchestration proved itself quickly and demand snowballed into hundreds of requests. Traditional automation could handle the predictable paths, but the real work is full of exceptions and judgment calls. The team’s approach was to get orchestration right first, then bring in AI agents that can reason and act where a person used to be the only option.
“We are highly regulated, so audit requirements are strict. Getting our processes into the platform, tracked and monitored, was what mattered first.”
What They’re Building
- A foundation first, intelligence on top. Greylock is taking the processes it already orchestrates and asking, process by process, where AI and agent-driven decisions can go further, starting with the ones ranked highest by time, volume, and member impact.
- Why Camunda. They need one platform spanning the full range, from deterministic straight-through steps to AI reasoning for the context-heavy cases, with the governance a regulated institution demands. It also lets them swap the underlying AI model as economics or performance change, without rebuilding the process.
- How it fits together. In the address-hygiene process, deterministic rules handle the routine while a large language model scores the exceptions. A person validates each result today, and the plan is to route trusted corrections to an AI agent, with consistent audit logging on every step. Capital BPM, Greylock’s implementation partner, helped accelerate the platform foundation this builds on, hitting an aggressive timeline.
“The AI scores every case, a person validates it, and as we get comfortable we let it take on more. That is how you bring AI into a regulated operation without giving up control of the process.”
“The insight that makes this work is architectural. A lot of processes are still deterministic, so you let rules handle that and bring AI in only where judgment is needed. Greylock saw that early and moved fast.”
What They’re Seeing So Far
Even at this early stage, the shift is showing up in two places: in what the orchestration layer has already delivered, and in what the first AI pilots are signaling.
Early signals
- A foundation that delivered. As part of an efficiency program focused on orchestrating more than a hundred processes, Greylock improved its efficiency ratio by nearly 7% and reached a three-year goal almost a year early.
- Problems they could not see before. Real-time visibility surfaced bottlenecks and issues the team did not know existed, even before optimizing individual processes.
What they expect at go-live
- Less manual review (projected). Once AI scoring is trusted, the plan is to let an agent make routine address corrections automatically, freeing people for the exceptions.
- A repeatable pattern (projected). The team expects to extend the same model, deterministic control plus AI plus human oversight, across its highest-impact processes.
“The exciting part is what is next. We are going to take our most important processes and go deep on where AI and agents can take them.”
What’s Next
The immediate focus is going deep on the processes that matter most, with address hygiene leading the way. It is going live with a person validating every AI result, and as confidence in the scoring grows, the process will route routine corrections to an agent. From there, Greylock plans to make agentic orchestration a repeatable pattern across the credit union: deterministic control where policy demands it, AI where judgment adds value, and a person in the loop wherever the stakes are high. The last few years built a foundation the team trusts. The next few are about building intelligence on top of it.
