Document Classification & Extraction
Accelerates invoice, KYC, contract and property-document processing for finance, compliance, legal and operations. Output is a structured schema with confidence, source coordinates and validation errors; low-confidence or high-risk fields always hand off to human verification.
| Document | Source | Predicted class | Class conf. | Extraction conf. | Validation | State | Action |
|---|---|---|---|---|---|---|---|
DOC-88231 · invoice_gulftech_0794.pdf GulfTech Facilities LLC · 3 pages |
Vendor portal | Vendor invoice | 0.99 | 0.94 | TOTAL_MISMATCH | Blocked — posting held | Review discrepancy |
DOC-88234 · tenancy_amina_noor_PR-A-101.pdf Amina Noor · Prototype Residences PR-A-101 |
Email intake | Tenancy contract | 0.97 | 0.98 | Pass | Extracted — auto-accepted | Open deal |
DOC-88240 · eid_front_masked.pdf KYC · Khalid Mansour (Falcon Trading FZ-LLC UBO) · 784-19●●-●●●●●●●-● |
KYC upload | Emirates ID | 0.98 | 0.78 | LOW_CONF: date_of_birth 0.61 | Needs verification | Send to maker-checker |
DOC-88244 · title_deed_PR-A-1104.pdf Noor Development Group · Building A |
Branch scan | Title deed | 0.95 | 0.88 | Pass | Needs verification — high-risk class | Verify |
DOC-88246 · pdc_cheque_00341.jpg Amina Noor · AED 42,000.00 · due 01 Sep 2026 |
Branch scan | Cheque / PDC | 0.92 | 0.83 | LOW_CONF: MICR line partial | Needs verification | PDC diary |
DOC-88249 · trade_license_falcon_fz.pdf Falcon Trading FZ-LLC · expiry check |
KYC upload | Trade licence | 0.96 | 0.97 | Pass | Verified | KYC profile |
DOC-88250 · offer_letter_PR-B-2207_v2.pdf Deal DL-2026-118 · Layla Hassan |
Email intake | Offer letter | 0.94 | 0.95 | Pass | Extracted — auto-accepted | Offer builder |
DOC-88252 · scan_20260717_0912.tif No template match · 1 page fax quality |
Branch scan | Unclassified | 0.44 | — | CLASS_BELOW_FLOOR (0.90) | Needs verification — manual triage | Triage |
DOC-88255 · utility_bill_dewa_jun26.pdf Address evidence · Amina Noor |
Tenant portal | Utility bill | 0.97 | 0.96 | Pass | Processing | — |
DOC-88257 · invoice_primelink_comm_0221.pdf Prime Link Realty · external broker commission |
Vendor portal | Vendor invoice | 0.98 | 0.91 | DUPLICATE_SUSPECT of INV-20260630-118 | Blocked — review | Anomaly alert |
| Field | Extracted value | Confidence | Source coordinates | Validation |
|---|---|---|---|---|
| supplier_name | GulfTech Facilities LLC | 0.99 | p1 · x120 y88 w340 h22 | Matched vendor V-0142 |
| supplier_trn | 100455623400003 | 0.98 | p1 · x120 y118 w210 h18 | TRN checksum OK |
| invoice_number | GT-INV-0794 | 0.99 | p1 · x520 y88 w130 h18 | Unique for vendor |
| invoice_date | 2026-07-14 | 0.97 | p1 · x520 y118 w110 h18 | Within open period |
| currency | AED | 0.99 | p1 · x520 y150 w40 h16 | Pass |
| subtotal | AED 45,000.00 | 0.96 | p3 · x510 y612 w120 h18 | Pass |
| vat_amount (5%) | AED 2,250.00 | 0.95 | p3 · x510 y640 w120 h18 | Rate matches config |
| total_amount | AED 47,250.00 | 0.94 | p3 · x510 y668 w130 h20 | AI-VAL-TOTAL-MISMATCH |
| line_items (12) | Σ AED 45,150.00 incl. VAT | 0.91 | p2–p3 · table region | Σ lines ≠ total (−2,100.00) |
| po_reference | PO-2026-0412 | 0.88 | p1 · x120 y150 w150 h16 | 3-way match pending — procurement |
idp-layoutlm-v3.4.1 · prompt PRM-KYCX n/a · 4.2 s · original file authoritative| Document | Reason for handoff | Assigned | SLA |
|---|---|---|---|
| DOC-88231 | AI-VAL-TOTAL-MISMATCH · posting blocked | DC Daniel Chen | 2h 48m left |
| DOC-88240 | date_of_birth 0.61 < 0.90 floor · Restricted-PII | YK Yusuf Karim | 3h 30m left |
| DOC-88244 | Title deed = always-verify class | EP Elena Petrova | 3h 55m left |
| DOC-88246 | MICR partial read · treasury custody | DC Daniel Chen | 3h 58m left |
| DOC-88252 | Class confidence 0.44 < 0.90 · manual triage | GO Grace Okafor | Overdue 22m |
Verification uses maker-checker: the maker confirms or corrects each flagged field against the source coordinates; a checker approves. Accepted corrections are written to the audit log with before/after values — see audit explorer.
- Extracted (0.61)
- 1988-01-07
- Corrected value
- 1988-07-01 source: p1 coordinates zoom
- Consequence
- Value flows to KYC profile draft only; original scan remains authoritative
| Class | Docs (7d) | Auto-accepted | Human-verified | Blocked | Avg confidence |
|---|---|---|---|---|---|
| Vendor invoice | 512 | 431 | 74 | 7 | 0.95 |
| KYC document | 366 | 0 (always verify) | 366 | 0 | 0.93 |
| Tenancy contract | 233 | 198 | 35 | 0 | 0.96 |
| Cheque / PDC | 158 | 0 (always verify) | 158 | 0 | 0.89 |
| Title deed | 67 | 0 (always verify) | 67 | 0 | 0.91 |
| Unclassified → triage | 34 | 0 | 34 | 0 | — |
| Total | 1,370 | 629 | 734 | 7 | 0.94 |
Static render mockup — no live data. Source: ERP Modern Development Specification v2.0 · Covers: AI-04 (structured extraction with confidence, coordinates, validation; low-confidence human handoff; posting blocked on total mismatch; audited corrections), AI-07 (no silent master-data update).