The Problem
A wealth advisory firm was running every client call through their phone system — and almost none of those calls were making it into the CRM. The vendor's add-on for call summaries existed but charged per-user, and the output quality wasn't trusted enough to push directly into the CRM. So advisors fell back on manual notes after the call, which meant either short notes (missing detail) or no notes (missing entirely).
Result: the CRM had a partial view of the client relationship. Whoever picked up the next call had to ask the client to repeat themselves.
What We Diagnosed
The vendor's add-on wasn't the right shape. We needed something that handled transcription, summarization, structured field extraction, and CRM upsert — all controllable, all auditable, and at a flat monthly fee instead of per-user.
We also identified what NOT to automate: judgment calls about suitability, advice quality, or compliance flags. The system summarizes and routes; humans still own the judgment.
What We Built
- Phone-system call log webhook ingestion for every advisor call
- Deepgram Nova-3 transcription
- Claude-powered summarization with structured field extraction (purpose, action items, follow-ups, sentiment indicators)
- Phone-number-to-contact matching with error handling for unmatched numbers
- Automatic CRM upsert with the summary, transcript link, and structured fields attached to the right contact
- Replaces the vendor add-on at a flat monthly fee regardless of advisor count
The Results
| Metric | Before | After |
|---|---|---|
| Advisor calls logged in the CRM | Partial | 100% |
| Time per advisor on post-call notes | 5–10 min | 0 min |
| Cost model | Per-user | Flat monthly |
| Continuity between advisor handoffs | Patchy | Full call context |
Every call captured. Every summary searchable. The CRM finally reflects the actual client relationship — not just the parts someone remembered to write down.