Skip to content
Under the hood

How Fieldiq extracts structured data at 99% accuracy

ML field detection trained per document class — not a generic OCR engine. Every extracted field carries an individual confidence score. Fields below your threshold are flagged and routed before they reach your ERP. This is what 99% field-level accuracy actually means.

See It On Your Documents View Pricing
Accuracy deep dive

Field-level extraction — not form-level OCR

Traditional OCR reads pixels as characters. Fieldiq's ML extraction understands what those characters mean in context — vendor ID, not just a string. Invoice number, not just a sequence of digits. That semantic understanding is why our field-level accuracy reaches 99%.

Document type classification
Automatically identifies whether an incoming document is an invoice, PO, claim, or filing — before field extraction begins.
Per-field confidence scoring
Every extracted field gets an individual confidence score. Fields below your threshold are flagged for exception — not silently passed through.
Multi-layout and handwriting support
Handles vendor-specific formats, hand-annotated documents, multi-page invoices, and scanned tiffs without per-vendor templates.
Structured output in JSON, CSV, or direct API push
Extracted fields map directly to your ERP schema. No intermediate cleaning step — data arrives in the format your system expects.
Sample extraction — Invoice #07421
INVOICE #07421 — Meridian Supply Co.
vendor_id: "meridian-supply" — conf 0.997
invoice_no: "07421" — conf 0.999
Date: 2026-06-01 — conf 0.994
total: 12480.00 — conf 0.998
tax_code: "TX-8500" — conf 0.991
due_date: "2026-07-01" — conf 0.996
6 fields extracted — zero exceptions — routed to ERP
Exception handling

When extraction flags a field — here's what happens

The 1% of documents that don't meet confidence thresholds go somewhere useful — not back to a BPO queue.

Step 1 — Threshold check
Confidence falls below your configured threshold
Each field is evaluated independently. A single low-confidence field triggers exception routing — other high-confidence fields continue to ERP.
Step 2 — Rule evaluation
Business logic checks run before routing
Amount vs approved vendor range. Due date in valid window. Tax code matching your jurisdiction table. Configured by your ops team, not hardcoded.
Step 3 — Routing to human reviewer
Right exception goes to the right person
Routing rules direct AP exceptions to AP team, claims exceptions to the claims adjuster queue. Not a generic inbox — configured by exception type and department.
Exception routing — review queue
EXCEPTION — invoice_07891.pdf
field: vendor_id — conf 0.72 — below threshold 0.90
routed → AP_TEAM — priority: standard
EXCEPTION — claim_04291.pdf
field: policy_no — rule: no_match_vendor_registry
routed → CLAIMS_ADJUSTER — priority: high
CLEAN — invoice_07892.pdf
7 fields — all conf > 0.95 — pushed to SAP
Output formats

Data in the format your system expects

Extracted fields push into SAP, NetSuite, Dynamics, or any ERP via REST API, webhook, SFTP drop, or structured JSON export. No middleware step between Fieldiq and your system of record.

Four ways data leaves Fieldiq

Use whichever pattern matches your infrastructure. Most teams use REST API push for their primary ERP and SFTP flat-file for archive and audit workflows.

  • REST API push — direct to SAP, NetSuite, Workday, Dynamics
  • Webhook delivery — triggers your existing workflow automation
  • Structured JSON — field-mapped to your schema definition
  • CSV flat-file — SFTP drop for batch processing workflows
JSON — extraction output 
// POST /api/v1/extractions/result
{
  "document_id": "inv_07421",
  "doc_type": "invoice",
  "status": "clean",
  "fields": {
    "vendor_id": "meridian-supply",
    "invoice_no": "07421",
    "total": 12480.00,
    "tax_code": "TX-8500",
    "due_date": "2026-07-01"
  },
  "confidence": 0.993,
  "exception": false,
  "erp_push": "SAP_S4HANA",
  "processed_ms": 847
}

See the extraction engine run on your actual documents.

Bring a sample invoice, claim, or PO to the demo call. We'll extract it live and show you the field-level confidence scores, exception routing logic, and ERP output format.

Book a live extraction demo Calculate your savings first