First Heritage has grown to roughly $838 million in assets, nine branches, and over 50,000 members. Growth of that kind widens the surface area of products, disclosures, and procedures frontline and contact-center staff are expected to know precisely and apply consistently.

Traditional search, static manuals, and LMS coursework did not surface answers fast enough in the moment of a member interaction. In the words of Zach Saunders, Director of Retail Operations & Training: “A lot of our stuff was scattered, folders here, the intranet there. Nothing was really searchable. I’d think, ‘I know that’s in a procedure,’ but finding it could take hours … people didn’t want to waste time digging, so they’d just ask the person next to them — that led to different answers depending on who you asked.”

The specific gap was instant, trustworthy, source-backed access to current policy and product knowledge, available in the flow of work and refreshed as the institution’s offerings change.

Because Knowledge Assistant consumes institutional documents — not transaction, account, or loan data — it required no integration with the core processor’s real-time framework. The integration layer is the institution’s policy repository and its identity system for role-based access. This document-centric nature is precisely why an internal knowledge assistant can be deployed quickly, at low integration risk, regardless of core.

The capability is proven: 21 community-institution adopters across 7 vendors. The differentiation is not whether an institution can stand one up — it is whether the build produces examiner-ready evidence as a byproduct.

Off-the-shelf models can parse PDFs and produce accurate, source-linked corpus responses. Retrieval-augmented generation and human guardrails add context to a base reasoning-LLM (Sonnet, GPT, etc.). Consequently, ground-truth collection — not finding a reliable LLM — is the biggest hurdle. But it is very much attainable.

First Heritage launched Knowledge Assistant in April 2024, rolling it out first to the contact center and then expanding across departments after rigorous internal testing. Pilot testing was adversarial by design — Saunders: “I tried to break it. We pushed hard on all the procedures, asking tons of questions, and ended up making changes to how some were written. It helped us improve our content just by showing us how the AI interpreted it.”

First Heritage has not published quantified performance metrics. What is publicly on record are operational outcomes described by the team running the tool day-to-day.

Zach Saunders, on the operational shift: “It’s just faster. Our frontline can get to answers instantly, and members don’t even realize they’re looking anything up. It’s made everything smoother — better turnaround times, better service. Calls don’t sit on hold as long. It’s noticeable.”

On onboarding: “We introduce new hires to Knowledge Assistant right away. It’s what they learn from the beginning, no outdated habits to unlearn … New hires lean on it constantly, and that encourages more seasoned employees to follow suit. The ripple effect is real.”