Regulated outbound contact center
≈ $40M revenue (approx.) · ~250 agents · three sites
Compliant voicemail lane beside the existing dialer
A voicemail lane leaves a message specific to each account, in a consented agent voice, at a volume no floor can match, and shares one attempt ledger with the dialer so contact limits stay enforced in one place.
- Organization
- Regulated outbound contact center
- Reality
- Most of an agent's day is answering machines
- Runs on
- The organization's own hardware
- Scope
- A compliant voicemail lane beside the existing dialer
Where they started.
Agents on a regulated outbound floor spend most of the day leaving voicemails. A person leaves a hundred rushed, identical messages; a robocall leaves a voice nobody returns. Hosted virtual-agent products would not share the organization's own contact limits, and the conversations are regulated, so the audio could not leave the building.
What was built.
- An open-weight speech model on the organization's own GPU speaks in consented agent voices, licensed in writing per team.
- Every message is specific to the account and follows the limited-content format the rules define, with the right callback line.
- Attempts feed the dialer's daily file, so contact caps are enforced in one place, and a return call reaches the team that left the message.
- The lane runs beside the existing dialer through its supported file and transfer surfaces; agents keep their tools.
What changed.
- Personal outreach at machine scale, inside the organization's own contact limits.
- One attempt ledger shared with the dialer instead of a second system to reconcile.
- No per-minute vendor bill, and the voices remain the organization's own assets.
- The same voice stack carries the organization's other phone workflows.
The conversations are regulated and sensitive. Keeping the voice loop and the transcripts inside the building keeps the organization's certifications intact.
On-premise hardware
Components involved.
More like this.
Community financial institution
≈ $2.4B in assets (approx.) · 30 branches · ~400 employees
- Use Case:
- Governed core-banking assistant
- Results:
- Staff ask questions of the core banking system through governed MCP tools. Anything consequential is composed and paused for a human, so the assistant is useful without being able to act alone.
Corporate back office
≈ $350M revenue (approx.) · PE-owned · six operating companies · a 12-person corporate team
- Use Case:
- Shared skills and brand assets portal
- Results:
- Invoice coding, on-brand presentations and monthly reports from one shared portal where the logic and the assets live in one place, instead of in each person's chat window.
Technology consultancy
≈ $8M revenue (approx.) · Boutique consultancy · ~25 consultants
- Use Case:
- Content production agent
- Results:
- Sales decks, one-pagers and campaign copy from a private agent with the brand system loaded, running on the company's own inference, with skills shared across the team rather than pasted between chats.
Have one like it?
Every spotlight started as a conversation about a process nobody liked doing, under a data-locality constraint.