How PO Guardian Works?
AI reads your documents. Rules and math validate them. You review what matters.
The Validation Pipeline
Document Arrives
- Your inbox is scanned automatically, or you upload a document directly.
- PO Guardian accepts purchase orders, fulfillment requests, and returns.
- PDF text is extracted instantly.
Classification & Extraction
- Document text is sent to OpenAI’s GPT-4o model for classification and field extraction.
- GPT reads the unstructured text and returns structured data — PO number, line items, prices, dates, contacts.
- This is the only step that uses a language model. All validation, scoring, and price checking is rule-based logic — no AI interpretation.
Price Verification
- Every line item is matched against your uploaded price list.
- Matching works across naming variations — prefix stripping, character substitution, partial matching.
- Prices are compared across all your tiers — Premium, Standard, Retail, or whatever tiers you use.
- Result: Verified Between Tiers Mismatch Not in Price List
- This is a direct database lookup. No AI involved.
Rule-Based Validation
- Fixed rules check every document the same way:
Same document. Same rules. Same result. Every time.
Scoring
- Every document starts at 100 safety, 100 confidence.
- Each issue deducts from a fixed penalty table:
| Issue | Safety | Confidence |
|---|---|---|
| Stale document (30–60 days) | −10 | — |
| Stale document (90+ days) | −20 | — |
| Ship date passed | −15 | — |
| Missing carrier | −3 | — |
| Price mismatch | −15 | −5 |
| Duplicate document | −25 | — |
| Math error | −10 | −10 |
- Safety = business risk. Confidence = extraction reliability.
- No hidden logic. Every deduction is visible on the document detail page.
Review & Act
- Flagged issues are presented with clear descriptions and recommended actions.
- Dismiss flags that don’t apply. Mark documents as reviewed.
- PO Guardian is advisory — it flags, you decide.
Advisory Only —
You Make The Call
PO Guardian does not process, approve, reject, or modify your documents.
It does not send replies, place orders, or trigger any downstream action.
Every flag is a recommendation. Every score is a reference point.
You review the findings, dismiss what doesn’t apply, and decide how to proceed.
Think of it as a second pair of eyes — not an autopilot.
Security & Privacy
Your Data Is Protected
- We connect via Gmail or Outlook OAuth — we never see or store your email password.
- We only read unread emails with attachments. We do not read, store, or access email body content.
- After processing an attachment, the email is marked as read. That’s the only change we make to your inbox.
- You can disconnect your email at any time from the dashboard.
- Uploaded documents are stored securely in an encrypted database.
- Document text is sent to OpenAI’s API for extraction — OpenAI does not use API data for training. See OpenAI’s API data usage policy.
- Your documents are never shared with other users or third parties.
- You can delete any document at any time.
- Your price list is stored in your account only.
- Price data is never sent to any AI model — verification is a direct database lookup.
- No other user can see your pricing.
- Vercel — enterprise-grade hosting.
- Supabase — database, encrypted at rest, SOC 2 compliant.
- Clerk — authentication, SOC 2 compliant, GDPR ready.
- Stripe — payments, PCI Level 1 certified.
- Delete any document, disconnect your email, or delete your entire account at any time.
- Read our full privacy policy at poguardian.com/privacy.
What makes this trustworthy?
Transparent Scoring
Every deduction follows a published penalty table. You see exactly why a score is what it is.
Your Data, Your Rules
Price verification uses YOUR price list. Validation uses fixed rules. Nothing is guessed.
Advisory, Not Automated
PO Guardian flags issues. It never processes, approves, or rejects documents on your behalf.
