Name: Invoice Data Extraction & Fraud Detection API | DVAP
Brand: datavance-1
SKU: api-datavance-1-dvap
Availability: InStock
Author: API Market

Amir Far

# DVAP - Invoice Data Extraction & Fraud Screening API

**Accounts payable you can hold accountable.** DVAP reads invoices the way your AP team does - every field, every line item - and hands back structured JSON with a confidence score on each extracted value, a validation verdict, and a fraud risk assessment with named, explainable reasons. No templates to configure, no per-vendor training, no documents stored.

DVAP is built and operated by DataVance (ABN 16 787 418 307), an Australian data-engineering business. It is used to automate invoice intake for finance teams, bookkeeping and BPO providers, and product teams embedding AP automation into their own software.

## Why teams pick DVAP

**Per-field confidence scores.** Most extraction APIs give you values; DVAP tells you how much to trust each one. Every field - invoice number, ABN, total, each line item - carries its own confidence score, so you can auto-approve what's certain and route only genuinely uncertain fields to a human. That single design decision is what turns "OCR output" into a straight-through-processing pipeline.

**Published, reproducible accuracy.** Accuracy isn't a marketing claim here: benchmark results are served live from the API itself (`GET /api/v1/metrics/accuracy`, no authentication required) and are versioned with the code that produced them. Every published number is reproducible from the exact commit that shipped it. You can check our accuracy before you send a single document.

**Cost-aware ingestion routing.** Digital PDFs, DOCX files, and plain text are parsed by direct text extraction - no OCR pass, no OCR latency. Scanned PDFs and images (PNG, JPG, TIFF) are automatically routed through AI-powered OCR. Every response reports which path was taken under `ingestion`, so you always know how a document was read.

**Stateless by design.** The endpoints on this listing store nothing. Your documents are processed in memory and the extraction is returned to you; there is no document retention, no training on your data, and nothing to delete later. That makes DVAP straightforward to approve for privacy-sensitive workloads.

## Endpoints

### `POST /api/v1/invoices/validate` - Extract & validate
Upload one document (multipart form-data, field `file`) and receive:

- **Extracted fields with per-field confidence**: invoice number, PO number, invoice date, due date, vendor name, vendor tax ID, ABN, currency, subtotal, tax, GST, VAT, total, discounts, freight, bank/payment details, payment terms, and a vendor industry classification.
- **Line items**: SKU, description, quantity, unit cost, tax, and total for every line, each with its own confidence score.
- **Validation errors**: arithmetic that doesn't add up, missing mandatory fields, malformed identifiers - the checks an AP clerk would run, done instantly.

### `POST /api/v1/invoices/fraud-check` - Score fraud risk
Upload one document and receive a fraud risk score (0-100) with named reasons, so every flag is explainable to an auditor:

| Signal | What it catches |
|---|---|
| `duplicate_invoice` | The same invoice submitted again |
| `bank_details_changed` | Payment details differ from what this vendor used before |
| `vendor_name_lookalike` | Vendor name imitating a known supplier |
| `totals_mismatch` | Line items that don't sum to the stated total |
| `currency_changed` | Vendor suddenly billing in a different currency |
| `due_before_invoice_date` | Due date earlier than the invoice date |
| `unseen_vendor` | First invoice ever seen from this vendor |
| `unusual_amount` | Amount far outside the vendor's normal range |
| `round_amount` | Suspiciously round totals, a classic fraud tell |

### `GET /api/v1/metrics/accuracy` - Published benchmarks
The measured extraction and fraud-detection accuracy of the exact code version currently serving your requests. Public, no key required.

## Quick start

```bash
curl -X POST "https://prod.api.market/api/v1/datavance-1/dvap/api/v1/invoices/validate" \
  -H "x-api-market-key: YOUR_API_KEY" \
  -F "file=@invoice.pdf"
```

Typical response (abridged):

```json
{
  "extracted": {
    "invoice_number": { "value": "INV-8203", "confidence": 0.99 },
    "po_number":      { "value": "PO-2026-0417", "confidence": 0.99 },
    "vendor_name":    { "value": "Cloudreach IT Services Pty Ltd", "confidence": 0.98 },
    "total":          { "value": 2376.0, "confidence": 0.99 },
    "payment_terms":  { "value": "Net 30", "confidence": 0.98 },
    "line_items": [
      { "sku": "MSP-STD", "description": "Managed IT support - July 2026",
        "quantity": 1, "unit_cost": 1450.0, "total": 1450.0, "confidence": 0.98 }
    ]
  },
  "extraction_confidence": 0.9,
  "validation_errors": []
}
```

One document per request; supported formats are PDF, DOCX, PNG, JPG, TIFF, and plain text. Malformed requests (missing file, empty body, oversized upload) return a 4xx status with an `{"error": "..."}` body describing what to fix - only successful processing returns 2xx, so you are never billed for a failed call. Authentication and quota issues use standard HTTP status codes.

## What you can build

- **Straight-through AP processing** - auto-post invoices whose fields all clear your confidence threshold; queue only the exceptions.
- **Pre-payment fraud gate** - screen every inbound invoice before it reaches your payment run; route flagged documents to a reviewer with reasons attached.
- **Bookkeeping automation** - turn client email inboxes full of PDFs into clean structured records at scale.
- **Data migration & audit** - batch-extract historical invoice archives into analyzable data, with confidence scores telling you which records need human review.
- **Embedded AP features** - add "upload an invoice" intelligence to your own product without building an extraction pipeline.

## Australian-ready, globally useful

DVAP extracts ABNs and GST natively alongside generic tax and VAT fields, understands `TAX INVOICE` conventions, and handles AUD as comfortably as USD, EUR, or GBP. Vendors and documents from any country work: field extraction is language-model driven, not template driven.

## Pricing, limits & support

Plans and rate limits are listed on this page's pricing section - pick the tier that matches your monthly document volume, and upgrade or downgrade at any time through API.market. Every request is metered by API.market; there are no side fees.

- **Website:** https://datavance.com.au
- **Terms & privacy:** https://datavance.com.au/legal/terms and https://datavance.com.au/legal/privacy
- **Support:** via the contact details on our website - we're a specialist team and provider questions land with an engineer, not a ticket queue.

## FAQ

**Do you store my documents?**

 No. The endpoints on this listing are stateless: documents are processed in memory, the structured result is returned to you, and nothing is retained or used for training.

**Do I need templates or per-vendor setup?**

 No. Extraction is model-driven, so a vendor DVAP has never seen extracts as well as a familiar one. Send your ugliest invoice first.

**What happens with multi-page or unusual documents?** 

Each request processes one invoice document. Multi-page single invoices are fine; the response's `ingestion` block reports pages read and whether OCR was used.

**How accurate is it really?** 

Don't take our word for it - call `GET /api/v1/metrics/accuracy` (no key needed) and read the measured benchmark results for the code version currently serving traffic.

Upload your first invoice and see the per-field confidence scores for yourself - the accuracy endpoint will tell you what to expect before you even subscribe.

Extract every invoice field with per-field confidence scores, validate totals, and score fraud risk with explainable reasons. No templates, nothing stored.

Invoice Data Extraction & Fraud Detection API | DVAP

BASIC