Skip to content

Confidence-Gated Document Automation

Messy Paperwork In. Automated API Callbacks Out.

Upload via Web UI or REST API. DocRelay extracts data from complex documents, judges accuracy field by field, and syncs verified records to your ERP — with zero manual entry. Deployed and tuned by iSwim's AI consulting team.

  • 20 sample docs
  • Results in 48 hours
  • NDA available
  • No commitment
Live resultcomplete

INV-2026-04417.pdf

Supplier invoice · EN / 繁中 · 1 chop

  • invoice_total

    HK$ 48,250.00

    99.2%
  • supplier_name

    示例貿易有限公司

    97.8%
  • company_chop

    Partially obscured

    81.4% · Needs review
POST https://erp.client.com/webhook→ 200 OK

How it works

Four steps. One clean callback.

Every field is checked before it reaches your system. Only the uncertain ones wait for a person.

  1. 01

    Upload (UI or API)

    Drag and drop in the web portal, or send POST /v1/documents.

  2. 02

    AI Data Recognition

    Fields, line-item tables, stamps and chops, handwriting. English, 繁中 and 简中 — even mixed on one page.

  3. 03

    Confidence Judgment

    Each field is scored against your thresholds (≥ 95%) and business rules, like totals and PO match.

  4. 04

    Intelligent Routing

    Path A — Confidence OK

    200 OK webhook with structured JSON to your system.

    Path B — Low confidence

    Only the uncertain fields wait for 1-click human review. Then the callback fires. Corrections improve the model.

For CTOs & integration leads

Clean JSON. No duplicates. Ready for your ERP.

One endpoint in. One webhook out. Every field carries its own confidence score, so your system always knows what was checked.

  • Send the same document twice and it is still processed only once — no duplicate records.
  • Works with any system that can receive a standard REST call.
  • Sandbox environment for UAT before you touch production.
POST /v1/documents
Authorization: Bearer sk_live_••••
Idempotency-Key: 7f3c…
Content-Type: multipart/form-data

POST https://erp.client.com/webhook → 200 OK

{
  "document_id": "doc_8Hk2…",
  "type": "bill_of_lading",
  "status": "auto_approved",
  "confidence": 0.982,
  "fields": {
    "container_no": { "value": "MSKU1234567", "confidence": 0.995 },
    "hs_code":      { "value": "8471.30",     "confidence": 0.971 },
    "gross_weight": { "value": 18450, "unit": "kg", "confidence": 0.988 }
  },
  "review": null
}

Industries

Built for the paperwork APAC actually runs on.

Bilingual forms. Stamped delivery notes. Handwritten receipts. Pick your industry.

Documents in

  • Bills of Lading
  • Air Waybills
  • Customs Declarations
  • Stamped Delivery Notes / POD

Data extracted

  • Container numbers
  • HS codes
  • Weights
  • Consignee / shipper
  • POD stamps and dates

Rules checked

  • Container number check digit (ISO 6346) is valid.

Callback destination

  • CargoWise
  • SAP
  • Custom WMS
Sample review case

pod_stamp_date: 1?/11/202678.6%

Stamp overlaps the handwritten date. Held for review. All other fields sent.

See the full use case: Logistics & Supply Chain

Documents in

  • Medical claim forms
  • Hospital receipts
  • Policy schedules
  • ID cards

Data extracted

  • Diagnosis codes
  • Itemised treatment costs
  • Policy numbers
  • Claimant IDs

Rules checked

  • Receipt total matches the claimed amount. Policy is active on the treatment date.

Callback destination

  • Claims management systems
Sample review case

receipt_total: HK$ 3,?80.0083.1%

Faded thermal receipt. One digit unclear. Reviewer confirms in one click.

See the full use case: Insurance

Documents in

  • Supplier invoices
  • Purchase orders
  • Bank statements
  • Remittance advice

Data extracted

  • Line items
  • Tax splits
  • Vendor details
  • Totals

Rules checked

  • 2/3-way match of invoice against PO (and GRN).

Callback destination

  • NetSuite
  • Oracle
  • Xero
  • SAP
Sample review case

line_items[3].qty: 12 vs PO 10rule fail

Quantity does not match the PO. Routed to AP with the PO line attached.

See the full use case: Finance & Accounting

AI Consulting

Software that ships with an AI consulting team.

iSwim's AI consultants work alongside your operations and IT teams, from first sample to production and beyond. You get outcomes, not a login.

  • Forward Deployed Engineer

    Works on site with your IT team. Builds the connection to your ERP, WMS or claims system, and stays until it runs smoothly in production.

  • AI Solutions Architect

    Designs the end-to-end flow, from document upload to the data landing in your systems.

  • Document AI Specialist

    Sets up the fields to extract, tunes accuracy on your real documents and decides when a person should check.

  • Business Process Analyst

    Maps how your team works today, the checks you rely on and where exceptions go. Logistics, insurance or finance.

  • Project Manager

    Runs the plan from sample test to go-live. One point of contact, weekly updates, no surprises.

  • Customer Success Lead

    Tracks accuracy and how many documents go through with no human touch. Runs monthly reviews to keep improving.

What the team delivers

  • Strong track record in enterprise automation across APAC
  • Bilingual team (EN / 繁中 / 简中)
  • Hong Kong-based
  • Process and document audit
  • ROI and business-case model for your CFO
  • Field schema and validation-rule design
  • Confidence-threshold tuning
  • ERP / WMS integration and UAT
  • Reviewer training and change management
  • Go-live hypercare
  • Ongoing model optimisation and quarterly business reviews
  • AI automation roadmap beyond documents (optional advisory)

Talk to an AI consultant

Engagement model

From sample to production in about two weeks.

  1. 1

    30 min

    Discovery Call

    We learn your documents, volumes, systems and success metrics.

  2. 2

    48 hrs

    48-Hour Sample Test

    You send 20 docs. You get a field-level accuracy report and a live callback demo.

  3. 3

    ~2 weeks

    Pilot & Integration

    Consultants configure rules and thresholds and connect your ERP or WMS in staging.

  4. 4

    ongoing

    Go-Live & Optimise

    Hypercare, live dashboards and monthly tuning.

Proven impact

Less keying. More done automatically.

reduction in manual processing time
90%
of documents processed fully automatically (no human touch)
86%
from sample test to live webhook integration
2 weeks
to your first field-level accuracy report
48 hrs

All figures from a logistics production deployment. Results vary by document quality and mix.

Security & integrations

Enterprise-ready from the first API call.

Developer-friendly API & webhooks

  • Clean JSON with per-field confidence
  • No duplicate records
  • Sandbox environment

Enterprise security

  • Encrypted in transit (TLS 1.3)
  • RBAC and SSO
  • Audit logs
  • Configurable data retention, including zero-retention options
  • Built with privacy in mind

Turnkey implementation by iSwim

  • Scoping, configuration and integration done for you
  • UAT and reviewer training
  • Hypercare and monthly tuning
Meet the AI consulting team

Sends verified data to

  • SAP
  • Oracle
  • NetSuite
  • Xero
  • CargoWise
  • Custom WMS
  • Claims platforms
  • Any REST endpoint

FAQ

Questions enterprise buyers ask.

What accuracy should we expect, and how is it measured?

It depends on your documents, so we measure it on yours. The 48-hour sample test gives you a field-level accuracy report: every field, its extracted value, its confidence score and whether it was correct. In production, a live dashboard tracks the same numbers every day.

What happens to documents the AI isn't sure about?

Only the uncertain fields are held. A reviewer sees the field next to the source image and confirms or corrects it in one click. Then the callback fires. Every correction is used to improve accuracy over time.

Do you handle Chinese, bilingual, handwritten or stamped documents?

Yes. DocRelay reads English, Traditional Chinese and Simplified Chinese, including mixed-language pages. It also reads handwriting and detects stamps and company chops. Hard cases are scored lower and routed to review instead of guessed.

How is our data protected, and is it used to train models?

Data is encrypted in transit (TLS 1.3). Retention is configurable, including zero-retention. Your documents are not used to train models for other customers.

Do we need templates or training data?

No. DocRelay understands document layouts from context, so new suppliers and formats work without templates. Our consultants configure field schemas and validation rules for your process.

How long does integration take, and what do we need from our IT team?

About two weeks from sample test to a live webhook in staging. Your IT team provides a webhook endpoint, test credentials and a contact for UAT. Our Forward Deployed Engineer handles the rest.

What does the AI consulting team do after go-live?

Hypercare first. Then monthly tuning reviews: thresholds, rules and new document types. Quarterly business reviews track how many documents are processed automatically, accuracy and savings against the business case.

How is DocRelay priced?

A fixed license fee plus a volume-based fee, scoped to your document volumes after the sample test.

Send us 20 sample documents. See live extraction and API callback accuracy in 48 hours.

Our AI consultants will walk you through the results — field by field.

Get a Free 48-Hour Sample Test

Tell us about your documents. We will send a secure upload link — no files are collected on this form.

Request type

Document types
Preferred workflow

We use these details only to reply to your request.