How AI matches emails to projects

The AI's job is to look at an incoming email and figure out: is this about an active project, and if so, which one? Here's the logic.

What the AI sees

For each email, the AI gets:

• The subject line
• The plain-text body (no images, no attachments)
• The sender's email address and name
• A list of all your active projects with: name, address, town, client name, and overview

How it matches

The AI uses several signals at once:

• Address match. If the email mentions "123 Main St" and one of your projects is at 123 Main St, that's a strong match.
• Client name match. If the sender or body mentions a client name on a project.
• Town/jurisdiction match. Useful when an inspector emails about a project in a specific town.
• Project name match. Direct mention of a project name in the subject or body.
• Topic match. The body content semantically matches what a project is about (this is fuzzy and used as a tiebreaker, not a primary signal).

What you see

Email Checker shows each processed email with the AI's suggested project match and proposed update — typically one of:

• Mark task X complete (with reasoning from the email)
• Add a note to task X
• Update the project status
• No change suggested (informational email)

You approve, edit, or reject. Approved updates apply immediately and the client portal reflects them in real time.

When the AI gets it wrong

Misses fall into a few categories:

• No match found: usually because the project description is too generic. Add specifics (address, jurisdiction, client name) to the project overview.
• Wrong project matched: usually because two projects share too many signals. The fix is the same — make project descriptions more specific.
• Wrong task identified: the AI picked the wrong task to update. Approve manually, the AI doesn't learn from individual rejections (yet) but does benefit from clearer task names.

Privacy note

Email content is sent to Anthropic's API. Anthropic's default policy is to not train on API submissions. We don't store full email bodies on our servers — only the matched project ID and the suggested update. The original email stays in your Gmail.