AI Transcription Summaries

Security aware meeting summaries

 Role:
Director of Product - Acted as sole PM

Timeline:
2.5 Months (Nov 2025 - Jan 2026)

Team:
7 Developers, UX Designer, R&D Prompt Engineering Support

The Challenge

Every Concierge customer call required a minimum of 2 engineers - one to engage with the customer, one dedicated to taking notes. This consumed over 40,000 hours annually across the team. When the SVP of Security Services pulled Zoom data showing the scale of this inefficiency, leadership prioritized finding a solution that would free capacity without sacrificing documentation quality.

The constraint:

We couldn't eliminate humans entirely. Engineers needed to remain involved to capture non-verbal context and ensure accuracy, but we could turn them from authors into editors.

We couldn’t use an off-the-shelf solution or Zoom AI summaries because the lack of security context and customization made the summaries useless.

My Approach

I scoped the solution to transform note-taking from manual authoring to AI-assisted editing. Working with our internal LLM team (trained on Arctic Wolf products and cybersecurity terminology for higher accuracy), we designed a workflow where engineers could download call transcripts, upload to Command Center, and generate structured summaries with one click.

Engineers could add notes for non-verbal context or emphasis, which fed into the prompt alongside the transcript. We piloted with select teams, setting clear expectations: review and edit AI output rather than write from scratch. This required quick iteration on prompts - handling edge cases like MSP calls (breaking out by end customer) and filtering inappropriate language.

The Solution

AI Transcription automates meeting documentation by processing call transcripts and engineer notes to generate structured summaries directly in Command Center.

Key capabilities:

One-click summary generation: Engineers upload transcripts and optional context notes. AI generates formatted summaries including attendees, discussion topics, security concerns, follow-up tasks, customer sentiment, and sales leads.

Integrated workflow: Summaries link directly to associated SPiDRs in Command Center, creating seamless documentation without context-switching.

Human-in-the-loop editing: Engineers review and refine AI output rather than authoring from scratch, maintaining quality while dramatically reducing time investment.

The Impact

Projected savings:

  • $1.6M annually in reduced headcount costs (41,600 hours saved across department)

  • 70% of meetings targeted to require only 1 engineer (down from 2) by eliminating dedicated note-taker role

Early validation:

  • >90% adoption since mandatory rollout in late January 2025

  • Accuracy feedback mostly positive with quick prompt iterations

Key Learnings

Speed mattered more than perfection. The projected savings ($1.6M annually) justified rapid iteration over exhaustive testing.

By scoping to "make engineers editors, not authors," we shipped in less than 3 months and achieved >90% adoption immediately. When ROI is clear and massive, optimize for learning velocity over polish.