Every investor email, a CRM record. Automatically.
A multi-agent AI pipeline that reads 300–500 investor emails a week, extracts who, what and why, and writes clean CRM records that always point back to the original message.
- 90% less manual work: analysts review records instead of retyping emails.
- 70+ hours back every week: time returned to the team for real investor work.
- Traceable by design: every field links to the sentence it came from.
Meet our client
- Client
- Investment data and intelligence firm
- Industry
- Financial data & investment intelligence
- Market
- Global
In a nutshell
The client manages relationships across a global network of investors, partners and companies, and most of that relationship lives in email. Analysts were reading every message and retyping it into the CRM. We built a multi-agent system on Anthropic Claude that ingests each email, extracts the people and organisations, works out what the message means, and writes a validated CRM record linked to its source.
The signals were there. Buried in the inbox.
Every week brought hundreds of emails carrying intros, commitments, declines and requests. Each one had to be read and retyped by an analyst. Set the weekly volume and watch the queue.
Illustrative: assumes about 12 minutes of analyst time per email, reduced by the measured 90%.
- Analyst time sunk in typing: reading, interpreting and re-entering every message by hand.
- A CRM that lagged: records were updated days after the conversation happened.
- Inconsistent capture: two analysts recorded the same kind of email in two different ways.
- No view of engagement: nobody could see which investors were warming up or cooling off.
Five agents. One email in, one clean record out.
Pick an email and run it through ingestion, extraction, classification, structuring and validation. Some emails make one record, some make three, and some rightly make none.
Every field points back to its sentence
Audit was a hard requirement, so nothing in the CRM floats free of its evidence. Click any field in the record to see exactly where it came from.
- Linked to the source: every record stores the message it came from, so anyone can open the original.
- Evidence per field: extracted values keep the passage that supports them.
- Validated before writing: required fields, known organisations and duplicates are checked first.
- Same rules every time: one schema for every email, whoever would once have typed it.
From emails to engagement trends
Once every email is a structured record, relationships become measurable. See who is warming up, who is cooling off, and the conversations behind each trend.
A pipeline that runs itself
n8n captures the mail, a FastAPI service coordinates the Claude agents, and PostgreSQL keeps every record with its source and audit trail, all in Docker containers in the cloud. Tap a component, or trace an email.
Messy inboxes, strict audit
Real investor email is informal, multi-topic and inconsistent, and the output had to stand up to audit. Flip each card from the challenge to how we solved it.
A manual workflow became an intelligence pipeline
Following production deployment, the firm processes its investor email automatically and sees its relationships in data, not just in inboxes.
Stop retyping. Start seeing.
AI agents that turn everyday email into structured, traceable data your team can act on.
- 01Your business objective
- 02The technical context
- 03The right next step