SOP: Extracting Knowledge from Emails/Unstructured Text¶
Context¶
We frequently receive high-value knowledge buried in unstructured formats like emails, Slack threads, or PDF attachments from vendors.
Do not simply save these files as PDFs in the Knowledge Base. A PDF is a "dead" artifact that AI cannot easily parse for specific answers, and humans cannot easily browse for quick facts.
The Rule: "Split the Atom"¶
You must separate the Information (Truth) from the Event (Record).
| Component | What it is | Where it goes |
|---|---|---|
| The Entity | Who is this about? (Vendor, Tool) | .../{system}/index.md |
| The Truth | Reusable facts (How-to, Pricing) | .../{system}/sops/ or reference/ |
| The Record | The actual conversation event | 90_records/correspondence/ |
Procedure¶
Step 1: Create/Update the Entity (The "Who")¶
Check if a folder exists for the vendor or subject. If not, create it.
* Location: 10_commercial/vendors/{vendor-name}/ or similar.
* Action: Update the index.md with high-level data found in the email (e.g., "Contact Person," "Account Number").
Step 2: Extract the "Evergreen" Content¶
Identify instructions that will remain true after today.
* Action: Create or update an SOP.
* Example: If the email says "To get the discount, click the button at the bottom," create sops/ordering-guide.md and write that step clearly.
* Constraint: Do not include "As we discussed yesterday" in this file. Write it as an instruction manual.
Step 3: Archive the Source (The "Record")¶
Save the original text so we have proof of what was promised.
* Location: 90_records/correspondence/YYYY-MM-DD-{subject}.md
* Action: Paste the raw email body here.
* Metadata: Link it to the Entity using the frontmatter field related_entity: "[[partner-name]]".
Example Scenario¶
Input:
Email from Vendor X: "Hey, we updated our API key to '12345' and you need to rotate it by Friday. Also, nice meeting you!"
Output:
1. Update: /01_architecture/catalogs/vendor-x-api.md (Update the key field).
2. Create: /20_operations/sops/sop-rotate-keys.md (If instructions were included).
3. Log: /90_records/correspondence/2025-11-26-vendor-x-update.md (Paste the full email).
Why we do this¶
- AI Accuracy: The AI can find the current API key in the Catalog without getting confused by old emails in the Records folder.
- Browsability: A human looking for "How to order" finds a clean guide, not a 30-message email thread.