Case study · Investor relations
Investor email to a reviewable CRM update in under five minutes.
At one private equity firm, investor-email triage went from days to under five minutes. A forward deployed engineer from By The People built a system on Claude that reads each flagged investor email, matches it to the records already in the firm’s CRM and proposes one set of updates, each value with its source passage. A person approves every change before it is written.
By Fatjon (Tony) Kalemaj, COO & Head of Applied AI · Published
The problem
Why did investor emails take days to reach the CRM?
During a fundraise, the firm’s investor relations team exchanges hundreds of emails with prospective limited partners. Each one carries something the CRM needs: a commitment moving stage, a new contact at a consultant, a meeting agreed for a roadshow, a document that has to be filed.
Turning those emails into CRM records was manual. Each message was read against a documented procedure and the records were updated by hand. The work was accurate but slow, and it did not keep pace with a live fundraise: at peak, information could sit for four to six days before it reached the CRM.
The firm had already established that Claude could read an investor email and interpret it correctly. What it did not have was a production system around that.
The build
What did the forward deployed engineer build?
By The People placed a forward deployed engineer inside the firm’s own environment to design and build the system. It runs Claude Opus 5 through the firm’s own deployment, so the emails stay inside its approved tools.
A language model cannot see a CRM, so it is never asked whether a record exists.
The system reads each flagged email, identifies the companies, contacts and funds it refers to, resolves them against the records already in the CRM, and proposes one consolidated set of updates. Every proposed value arrives with a confidence level and the passage it came from, so a reviewer can see why the system suggested it.
Claude interprets
Reads each flagged email, identifies the companies, contacts and funds in it, and chooses between the candidate records it is shown.Code decides
Looks up the records already in the CRM, holds the confidence thresholds and enforces the rules on what may be written.A person approves
Sees the fields the system is uncertain about, approves the rest, and signs off every change-set before it is written.
The results
How was the result measured?
The figure is the time from a flagged investor email to a reviewable set of proposed CRM updates. At peak fundraising that used to be as long as four to six days, the time information sat before it reached the CRM. It is now under five minutes.
The system went live in September 2026 and has processed several hundred investor emails, the investor email flow of a live fundraise.
| Before | With the system | |
|---|---|---|
| Flagged email to CRM update, at peak | As long as four to six days | Under five minutes to a reviewable set of updates |
| Who reads the email | A person, against a documented procedure | Claude, which identifies the companies, contacts and funds |
| Whether a record exists | Checked by hand | Looked up by code; Claude chooses between the candidates |
| What the reviewer sees | The whole email, retyped into the record | The uncertain fields, each with its confidence and source passage |
| What reaches the CRM | Whatever was typed | Only a change-set a person has approved |
What changed
What does the investor relations team do differently now?
The reviewer’s job has moved from transcription to judgment. They see the fields the system is uncertain about and approve the rest, and nothing reaches the CRM until a person has signed off the change-set.
Removing the reviewer was never the goal. The next phase runs in the same direction: widening the range of decisions the system can propose with confidence, so that a person’s attention goes only where it is needed.
The reviewer used to retype the email. Now they decide the fields the system is unsure of.
Questions on the investor-email triage
Does a person still approve the CRM updates?
Yes. A person approves every change-set before anything is written to the CRM. The system shows the fields it is uncertain about, each with its confidence level and the passage it came from, and the reviewer approves the rest.
Which AI model does the triage run on?
Claude Opus 5, through the firm’s own deployment, so investor emails stay inside the firm’s approved tools under its own data terms.
Why is the model not asked whether a CRM record exists?
Because a language model cannot see a CRM. The application looks up the records already there and Claude chooses between the candidates it is shown. The lookup, the confidence thresholds and the rules on what may be written are all code.
Sources
- By The People Technologies, investor-email triage engagement: the client-approved account of the build and its results (October 2026).
Have an inbox like this one?
Tell us where the work sits today and what it is written into. We will tell you whether a system like this fits it, and say so if it does not.