Alloy Labs' 7-Person Non-Technical Team Runs 25 Meetings a Week and Owns the Code

Alloy Labs Alliance is a peer consortium of nearly 90 US community and mid-sized banks, run by a seven-person non-technical operating team. The same team now runs 25 meetings a week through an automated pipeline and owns the code that drives it.

By early 2026 the team was triaging every transcript by hand into Zoho, Notion, and Slack. Sales calls, member check-ins, Center of Excellence sessions, and partner introductions, each producing institutional knowledge worth preserving.

The original brief called for n8n, a popular no-code automation tool. We recommended a custom application instead, sized for the team to extend on their own with Claude Code as their co-pilot.

In its first six weeks live, the system processed 150 transcripts, wrote 485 insights to Notion, and posted 127 notes to the right Zoho records. Within a month of handoff, the project lead was modifying the codebase herself.

The Challenge

Alloy Labs Alliance is a peer consortium of nearly 90 US community and mid-sized banks, representing roughly $500 billion in combined assets. A seven-person operating team runs Centers of Excellence (multi-bank working groups on topics like AI, payments, and lending), member onboarding, partner introductions with portfolio companies, and internal strategy. By early 2026 the team was running roughly 25 meetings per week, and every one of them produced institutional knowledge worth preserving: which bank said what, which partner committed to what follow-up, which insight should land in the consortium's knowledge base.

The team was working through transcripts with Claude's help and copying the right pieces into Zoho CRM, Notion, and Slack by hand. The work was getting done. Barely. Notes lagged, drifted, or never landed. Insights that should have compounded across the membership stayed locked inside individual transcripts. The bottleneck was not strategy. The team had already written a 28-page brief spelling out exactly what they wanted. The bottleneck was capacity.

The Complications

The team faced three escalating challenges that made their existing way of working untenable.

The non-technical-team paradox. Madeline Fredin, the project lead who authored the brief, is the most technical person on staff. Even she described herself as a Claude power user, not a developer. Nobody on the team writes code for a living. Yet every week they were summarizing transcripts and shuttling notes by hand into three different systems.

Center of Excellence insights were not compounding. Centers of Excellence are where the consortium creates the most value: multiple member banks in one room, comparing notes on a shared topic. But the existing workflow either copied a generic summary into every participating bank's CRM record (noise) or skipped CRM entry altogether. The brief called this out as critical. Each bank needs the comments and insights specific to it, never a shared blob.

The path they assumed they needed would have created the dependency they feared. The original brief specified n8n, the obvious no-code path for a non-technical team. But on the kickoff call, Madeline raised the deeper concern: "I'm a little bit concerned that if we set up a custom code base it's like we don't have anybody on staff who can maintain it. So we're just potentially in a situation where we're going to be reliant on you."

The Solution

We started by going back to the brief and stress-testing its load-bearing assumption. The team had picked n8n because it looked like the path a non-technical operator could maintain on their own. But that assumption belonged to 2021. In 2026, a non-technical team's most accessible engineer is an AI coding agent like Claude Code. Those agents work better with real code than with the visual workflow diagrams n8n produces. The "safe" choice in the brief was actually the long-term trap.

What we recommended instead was a custom application, hosted in the cloud, sized for the team to extend over time with Claude Code as their partner. It came with a maintenance plan, not just a build plan: a default "dry run" mode so nothing is sent to live systems until the team flips a switch, alerts in Slack whenever something needs human review, and a handoff document written specifically for a client operator and the coding agent helping them run or modify the project locally. When Madeline asked who would maintain the system when things needed changing, the answer was: Claude Code, with you in the driver's seat.

From there, the build focused on three things the team had asked for and one they hadn't.

The system reads every transcript and sorts it into one of six categories: Sales Call, Member Onboarding, Center of Excellence, Partner, Internal, or "Other" for anything it isn't confident about. When the system isn't sure, the meeting routes to a Slack review queue instead of being filed in the wrong place. Member-specific notes land on the right Zoho record. Internal-only meetings stay in Slack. Short, evidence-backed insights land in Notion as searchable rows.

For Centers of Excellence, the system does something the brief called critical. Instead of posting one generic summary across every bank that attended, it isolates the comments specific to each bank and writes a separate, targeted note onto only that bank's CRM record. Banks that attended but did not say anything substantive get no entry. That is the quality bar a consortium needs.

The piece the team had not asked for was the safety net. When a meeting cannot be matched cleanly to an existing CRM record, the system does not fail silently. It surfaces the unresolved meeting to Slack with the note content attached, ready for a one-click replay. Insights without strong evidence get skipped rather than fabricated. Ambiguous bank names get flagged for human review.

Within three weeks of the kickoff call, the system was live. Before the formal handoff was even complete, Madeline was inside Claude Code herself, prototyping the next layer of her broader growth-engine design: turning the new observation stream into weekly trend briefs the consortium could publish.

Quantified Impact

In its first six weeks of live operation (April 13 to May 24, 2026), the system processed 150 meeting transcripts with a 94% end-to-end success rate, wrote 485 evidence-backed insights to the Notion knowledge base, and posted 127 notes onto the right Zoho records. Throughput averaged 25 meetings per week, peaking at 30 in the most recent week. The system extracted insight candidates at a rate of 3.4 per meeting; the evidence filter caught and dropped the ones without strong enough support before they reached Notion.

All six designed meeting types are running in production. The per-bank Center of Excellence routing is working as designed. The exception queue surfaces unresolved cases in Slack rather than dropping them.

The harder-to-measure shift is who is holding the system. Madeline reviews the Slack alerts, cleans up the rare Zoho record that needs a bank name correction, and is modifying the codebase using Claude Code roughly a month after the official handoff. For a seven-person team supporting a 90-bank relationship base, that is the real shift. The system does not just save time. It frees the project lead to build the next thing.