# Inferio — full reference

> Inferio is an AI-powered OCR platform that turns invoices, receipts, bank statements, contracts, and KYC documents into validated, system-ready data. Japan-compliant out of the box (T-number + Electronic Bookkeeping Law). Built for AP, compliance, and integration teams. This is the comprehensive AI-readable reference; the shorter digest is at https://inferio.xyz/llms.txt.

Canonical site: https://inferio.xyz · Developer docs: https://docs.inferio.xyz · Built by LAR Technology.

## What it is

Four compounding problems hit every AP, compliance, and integrations team when document volume grows past a few hundred per month.

## Key facts

- **Category:** AI-powered OCR / intelligent document processing for invoices, receipts, bank statements, contracts, and KYC / identity documents.
- **Extraction:** Layout-aware field extraction powered by Claude Vision — vendor, T-number, line items, tax breakdown, dates, and amounts — not template-bound positional OCR.
- **Japan compliance:** Built for the Qualified Invoice System — T-numbers are validated in real time against the National Tax Agency (NTA) public API. Documents are archived with cryptographic timestamps to meet Electronic Bookkeeping Law (電子帳簿保存法) requirements.
- **Validation & accuracy:** Domain-specific validators (T-number, IBAN/BIC, date/amount consistency) run on every document; low-confidence fields (< 0.75) surface in a human-in-the-loop correction UI, and corrections feed back into the model.
- **Integrations:** Direct sync to freee, MoneyForward, Salesforce, NetSuite, SAP, or a custom ERP. Core REST API (idempotent, retry-safe, batch-friendly) plus Node / Python / Go SDKs.
- **Inputs:** PDF / PNG / JPG / TIFF via drag-drop, email webhook, or REST API. Multi-page and batch documents supported.
- **Pricing:** Pay per page. No setup fee, no subscription, no charge on rejected pages. Trial credits cover the first batches; the per-page rate falls automatically with monthly volume, down to about $0.005 per page at scale. Enterprise volumes get a custom rate card.
- **Enterprise:** Inferio Enterprise handles B2B document processing for banking, logistics, and large SMEs (trade invoices, letters of credit, KYC), with a paid proof-of-concept on your own documents.

## Why teams use it

- **Japan compliance, built-in** — Real-time T-number verification via NTA public API. Automatic 8% / 10% tax-rate classification per line item. Archival with cryptographic timestamps — Electronic Bookkeeping Law compliant by default.
- **Direct system sync** — freee & MoneyForward OAuth integration — extracted invoices land as journal entries in real time. ERP sync via REST API + webhook. Optional white-label for system integrators.
- **Human-in-the-loop** — Per-field confidence scores. Anything below 0.75 surfaces in a correction UI before sync — operators clear exceptions only, not every field. Corrections feed back as fine-tuning signals.
- **Production-validated** — Built by an AI engineering team with banking LC verification, payroll OCR (10k+ docs/month), and multi-agent document routing already in production. REST + webhook from day one.

## Manual document processing breaks at scale

Four compounding problems hit every AP, compliance, and integrations team when document volume grows past a few hundred per month.

- Speed & scale ceiling → Manual entry caps at ~500 docs/month per FTE. Inferio processes a single page in <5s and scales linearly with no human bottleneck.
- Error rate & tax risk → Human entry runs 1–3% errors. One bad digit on a JP T-number invalidates tax deductibility. Inferio validates every T-number against the NTA public API — zero spot-checks.
- Regulatory compliance gaps → JP Qualified Invoice System + Electronic Bookkeeping Law require precise field extraction and 7-year structured storage. Inferio's archival layer is compliant by default.
- System silos & re-entry → Extracted data syncs directly to freee, MoneyForward, or any ERP (SAP, NetSuite, Salesforce, in-house) via OAuth + webhook. No CSV gymnastics, no double entry.
- Off-the-shelf OCR fails on JP mixed-script and low-quality scans → Claude Vision is layout-aware and reasons about context — 99%+ accuracy on JP invoices, including handwritten fields and rotated phone shots.
- Black-box accuracy with no recourse → Per-field confidence scores. Anything below 0.75 surfaces in the correction UI before sync — operators clear exceptions only, not every field.
- Custom internal pipelines are expensive to build and maintain → REST + webhook integration ships in days, not quarters. No ML infrastructure to own. SLA-backed uptime.

## Ways to use Inferio

- **Web dashboard** — Drag-drop supplier invoices into the Admin UI. Approve or correct exceptions before sync. T-number validation, freee + MoneyForward auto-journals, full audit trail — no code required.
- **Inferio Enterprise** — B2B document processing for banking, logistics, and large SMEs. Trade invoices, letters of credit, bank statements, payroll records, KYC docs — all on the same engine. ERP integration via REST + webhook.
- **Core REST API** — Plain HTTP + JSON. Idempotent, retry-safe, batch-friendly. Drop the same engine into your own product or AP automation pipeline — no middleware required.

## How document processing works

1. **Ingestion** — PDF / PNG / JPG / TIFF via drag-drop, email webhook, or REST API. Multi-page documents and scanned images handled with automatic format detection.
2. **Pre-processing** — Auto-deskew, noise reduction, contrast enhancement, and upscaling for low-quality scans. Consistent normalised input regardless of source quality.
3. **AI extraction** — Claude Vision performs layout-aware field extraction: vendor, T-number, line items, tax breakdown, totals. Structured JSON with per-field confidence scores.
4. **Validation** — Domain-specific validators: T-number via NTA public API, IBAN/BIC for banking, date and amount cross-checks. Anomalies flagged for human review.
5. **Human-in-the-loop** — Low-confidence fields (< 0.75) surface in a lightweight correction UI. Corrections feed back as fine-tuning signals — accuracy improves with usage.
6. **System integration** — Validated data syncs to freee, MoneyForward, Salesforce, NetSuite, SAP, or custom ERP via REST + webhook. Full audit trail maintained per document.
7. **Compliant archival** — Documents archived with cryptographic timestamps (Electronic Bookkeeping Law compliant). Metadata indexed for full-text search. 7-year retention by default.

Need on-prem model deployment for maximum data isolation? Available for enterprise contracts — ask about the SOC 2 path during the discovery call.

## AP automation, measured in hours not days

A representative AP team running 2,000 invoices a month. Numbers from a benchmarked deployment — your scope mileage varies.

- **267 → 10 hours / month** — Staff hours drop from full data-entry workload to exception review only. 90%+ of documents flow straight through to ERP without human touch.
- **$6,675 → $850 / month** — Labor at $25/hr collapses from $6,675 to $250 in exception review, plus ~$600 in Inferio platform cost. Net $5,825 saved every month.
- **1–3% → 0% T-number errors** — Manual spot-checks miss invalid registrations. Every T-number is validated against the NTA public API — invalid invoices block before they hit your books.
- **Days → hours approval cycle** — Extracted data lands in your ERP the moment the page completes. Procurement reviews against the source PDF in the same screen; no re-keying, no waiting.

ROI illustration assumes a mid-size AP operation; client-specific savings are modelled during the discovery call against your actual document mix.

## Built for documents that matter

Tax invoices, ID documents, letters of credit: sensitive by default. Four guarantees that hold when they land on our infrastructure.

- **Encrypted in transit and at rest** — TLS 1.3 on every request. AES-256 on every object stored to AWS S3. Per-tenant data isolation — no cross-merchant access at any layer.
- **Documents stay yours** — Originals live on your infrastructure (or our short-retention staging bucket on hosted plans). Inferio processes in-memory only. AI provider does not retain or train on your documents.
- **Full audit trail** — Every extraction, validation, correction, and sync logged with timestamp. Exportable per request. Electronic-Bookkeeping-Law-grade evidence chain by default.
- **On-prem option for enterprise** — Need maximum data isolation? Enterprise contracts can deploy the model on your own infrastructure. SOC 2 Type II certification on the roadmap as we scale enterprise contracts.

## Seven document types. Four integration surfaces.

Pick a built-in template (each with a pre-tuned JSON schema) or define your own. The matrix below shows where each document type ships pre-wired across the integration surfaces.

Destinations: freee, MoneyForward, REST API, Custom ERP.

Sources: Trade invoice (JP) (T-number + 8/10% tax); Letter of credit (40+ fields incl. Incoterms); Bank statement (IBAN/BIC + balances); Payroll record (Salary + deductions + tax); KYC / ID doc (MRZ + identity fields); Receipt / expense (Merchant + tax + category); Custom schema (Bring your own JSON shape).

Seven document types ship pre-tuned. For anything else — purchase orders, lab reports, court filings — define a JSON schema in the dashboard and the same extractor adapts without retraining.

## Pay per page. Rate drops with volume.

No setup fee, no subscription, no charge on rejected pages. Trial credits cover your first batches.

$0.005 per page at scale · rate drops automatically with monthly volume.

Per-page rate ladder:

- Pilot · < 1,000 pages / mo — $0.012 / pg
- Growth · 1,000 – 10,000 pages — $0.008 / pg
- Scale · 10,000 – 100,000 pages — $0.005 / pg
- Enterprise · > 100,000 pages — Custom

Volume tiers apply automatically against your rolling 30-day usage. Failed pages (unreadable scan, low confidence) are never billed.

No setup fee; Trial credits on signup; No charge on rejected pages.

Every page includes:
- Trial credits cover your first hundreds of pages
- Real-time T-number validation against NTA public API
- freee + MoneyForward OAuth sync · auto-journal entries
- Automatic 8% / 10% tax-rate classification per line item
- Electronic Bookkeeping Law-compliant archival with cryptographic timestamps · 7-year retention
- Correction UI · confidence scoring · audit trail per document

Enterprise · banking · logistics · large SMEs Paid POC, then per-document or seat license. Start with a 2-week paid POC ($3,200 fixed) — live integration on your own sample documents with a published accuracy benchmark report. Production: $0.03 – $0.20 / doc (volume-tiered) or a monthly seat license. White-label available for system integrators.

Pricing: https://inferio.xyz/pricing

## Drop OCR into your backend in an afternoon

Bearer-token auth. Upload sessions for batch flows. Signed webhooks fire the moment each page is extracted. Idempotency keys make every retry safe.

- **API tokens** scoped per merchant + environment (test / live). Generate, rotate, and revoke from the dashboard. Server-side only — never embed in a browser bundle.
- **Upload sessions** Each batch gets its own session_id with per-file size + count limits. Validation happens client-side so unreadable scans never spend a credit.
- **Signed webhooks** Every extraction completes with a HMAC-signed POST. Dead-letter queue replays anything you miss. ocr.document.completed is the canonical fulfillment event.

## Online OCR — extract data from invoices & receipts

Drop a photo of an invoice or receipt below and watch Inferio's AI extract structured fields in seconds — no signup, no credit card. This is the same extraction engine that runs behind our API.

OCR tool: https://inferio.xyz/tools/ocr-demo

## Partner program

Inferio ships an embedded partner program for the channels closest to our buyers: system integrators bidding on AP automation projects, accounting firms (税理士事務所 / tax accounting firms) advising SME clients, and ERP consultancies wiring up document workflows for enterprise customers.

- **Apply for the program** — One form. We approve within a business day and mint your partner code + co-marketing kit.
- **Bring your clients on** — Embed your partner code in onboarding flows, agency pitches, or accountant newsletters. Referred customers attach automatically.
- **Earn monthly recurring** — Base commission on every referred customer's monthly bill — plus tier boost as you scale. White-label revenue share available for system integrators.

Tiers reward partners who scale — More referrals, bigger boost. Tier perks stack on top of base commission — Slack channel, quarterly bonuses, branded landing pages.

- **Starter** — 0 referrals; Base commission; Monthly payout · dashboard analytics
- **Silver** — 10 referrals; Base + 1% boost; Priority support · co-marketing assets
- **Gold** — 50 referrals; Base + 3% boost; Slack channel · quarterly bonus pool
- **Diamond** — 200 referrals; Custom rate card; Dedicated AM · branded landing page · white-label option

## Frequently asked questions

### Is Inferio compliant with Japan's Invoice System (the Qualified Invoice System)?

Yes — that's the headline use case. Every extracted T-number is validated against the National Tax Agency public API in real time, with the response (valid / invalid / not-found) stored on the document record. Tax-rate classification (8% reduced vs. 10% standard) runs per line item. The system surfaces invalid registrations before they hit your accounting books.

### Does the archival meet the Electronic Bookkeeping Law requirements?

Yes. Documents are archived with cryptographic timestamps, metadata is indexed for full-text search (vendor, date, amount, document type), and the default retention period is 7 years. The audit trail captures every extraction, validation action, correction, and sync — exportable per request as an Electronic-Bookkeeping-Law-grade evidence chain.

### Which accounting systems sync directly?

freee and MoneyForward via direct OAuth 2.0 API integration — extracted invoices land as journal entries in real time. For ERP systems (SAP, NetSuite, Salesforce, in-house) we integrate via REST + webhook. Custom downstream pipelines wire up the same way — sign the webhook receiver and we'll handle delivery + retries.

### What documents can Inferio handle today?

Seven document types ship pre-tuned with their own JSON schemas: JP trade invoices (with T-number), letters of credit (40+ fields including Incoterms), bank statements (IBAN/BIC + balances), payroll records (salary + deductions + tax), KYC / identity docs (MRZ + fields), receipts, and purchase orders. For anything else, define a JSON schema in the dashboard and the extractor adapts without retraining.

### Can Inferio read handwriting?

Yes — the vision-language engine reads handwriting in context (a handwritten total on a printed invoice, amounts and dates on forms, short notes), which traditional character-matching OCR could not do. Handwriting accuracy is always lower than printed text, so every extracted field carries a confidence score and anything below the 0.75 threshold routes to human review instead of posting silently.

### Does it handle scanned PDFs, multi-page files, and phone photos?

Yes. The product ingests PDF, PNG, JPG, and TIFF including multi-page documents — a scanned PDF is read page by page the same way as a photographed receipt. Phone photos are pre-processed (deskew, noise reduction, contrast) before extraction, so a slightly tilted shot taken at a desk is normal input, not an edge case.

### Can the extracted data feed my own AI or analytics?

That's the point of structured extraction: every document becomes queryable fields (vendor, dates, line items, tax, totals) instead of a dead image. Pull it through the REST API or signed webhooks into your database or warehouse, and it becomes the data foundation your internal analytics, reconciliation, or AI assistant answers from.

### How accurate is it on real Japanese documents?

99%+ field-level accuracy on JP mixed-script invoices and 99.2% on letter-of-credit fields in our banking deployment. Claude Vision is layout-aware and reasons about context, so it handles handwriting, rotated phone shots, and noisy scans far better than traditional OCR. Fields below 0.75 confidence are flagged for human review — we treat that as a feature, not a failure mode.

### How does the enterprise paid POC work?

Fixed fee — $3,200 / ¥500,000 for 2 weeks. We run live extraction on 3–5 of your real document types in a sandboxed environment, benchmark per-field precision against your ground truth, connect to your test accounting / ERP system, and deliver a written accuracy report. If the benchmark hits your targets, we agree on production scope; if not, you've spent two weeks of fixed-fee learning.

### What's the cost per page at volume?

Pay-as-you-go per processed page. Public ladder: $0.012 / page under 1,000 pages / month → $0.008 from 1,000–10,000 → $0.005 from 10,000–100,000. Past 100,000 pages / month customers qualify for enterprise per-document pricing ($0.03–$0.20 per doc depending on type complexity + SLA) or a monthly seat license for predictable cost.

### Where do my documents live?

Originals stay on your infrastructure by default. Inferio processes them in-memory only — the AI provider does not retain or train on your documents. Hosted plans have an optional short-retention staging bucket (configurable, default 7 days) for replay. Enterprise contracts can deploy the model on-premise for maximum data isolation.

### Do you have any production deployments to point at?

Yes. We've shipped an end-to-end letter-of-credit extraction system for a major financial institution (40+ fields, 99.2% accuracy, replaced a 6-person manual review team), a payroll OCR platform processing 10,000+ documents / month with direct accounting integration, a RAG-based enterprise support bot deflecting 78% of queries from human agents, and a multi-agent document router classifying 15 document types and dispatching to downstream workflows.

### Is there a Node / Python / Go SDK?

Not yet — but the REST API is plain HTTP + JSON, so `axios.post` / `requests.post` / `http.Post` works today. We're shipping idiomatic SDKs for Node, Python, and Go next; the proposal stack uses Python / FastAPI on the server side so a Python SDK lands first. Drop your email in the Developers section to get the beta invite.

### What does implementation look like?

Five phases over ~6 weeks for a standard rollout. Phase 1–2 (weeks 1–3): core pipeline + T-number validation + Electronic-Bookkeeping-Law archival on your tenant. Phase 3 (weeks 4–5): freee / MoneyForward (or your ERP) integration + correction UI wired into your AP workflow. Phase 4 (week 6): full QA cycle, security review, audit-trail walkthrough. Phase 5 (post-launch): multi-doc-type expansion (LC, payroll, KYC) as new document classes come online.

### How does the partner program work?

System integrators, accounting firms, and ERP consultancies apply for a partner code. When referred customers sign up and start processing, you earn a base commission on their monthly bill plus a tier boost (Silver → Diamond) as you bring more on. White-label revenue share available for system integrators bidding on AP automation projects.

## Guides on automating documents with AI OCR

Practical guides on AI-powered OCR, template vs. freeform extraction, and accounts-payable automation — for teams reading in English, Japanese and Vietnamese.

- [What is AI OCR (and how is it different from traditional OCR)?](https://inferio.xyz/blog/what-is-ai-ocr) — AI OCR uses a vision-language model to read a document the way a person would — understanding layout, labels, and context — instead of matching pixels to a fixed template. That's why it can extract data from an invoice it has never seen before, in a currency and language it wasn't specifically trained on, without an engineer pre-mapping every field.
- [Why AI is only as accurate as your data (and why OCR comes first)](https://inferio.xyz/blog/ai-needs-your-data) — An AI model is only accurate about your business when it can see your business's data. A general-purpose model knows nothing about your vendors, your contracts, or last month's spend — that knowledge lives in your invoices, receipts, and agreements, most of it locked in paper and PDFs. Building the data foundation that makes AI useful therefore starts with one unglamorous step: extracting structured data from documents. That step is OCR.
- [Template vs AI freeform extraction: which one for your documents?](https://inferio.xyz/blog/template-vs-freeform-extraction) — Template extraction uses a pre-tuned schema for a known document type and gets you slightly higher precision and speed; freeform extraction lets the model infer structure from any layout it has never seen. Most teams end up using both — templates for their highest-volume, most standardized documents, freeform for everything long-tail — and Inferio's "from sample" flow lets you turn a real document into a new template in minutes instead of writing one by hand.
- [How do you turn a photo into structured data, not just raw text?](https://inferio.xyz/blog/image-to-structured-data) — You turn a photo into structured data by running it through a vision-language model that reads the document's layout — not just its characters — and returns labeled fields like vendor, date, total, and line items as JSON, with a confidence score on every field. The difference from ordinary "image to text" is that the output goes straight into Excel or accounting software without anyone re-typing it.
- [How do you extract data from a PDF (digital or scanned)?](https://inferio.xyz/blog/extract-data-from-pdf) — To extract data from a PDF reliably, you need two steps: read the content of every page (whether it's a digital PDF with a text layer or a scanned image), then map that content to named fields — invoice number, total, tax, line items — with a confidence score on each value. Copy-paste and generic PDF-to-Excel converters only attempt the first step, which is why they break the moment a table spans pages or a new vendor sends a different layout.
- [Can OCR read handwriting? An honest answer](https://inferio.xyz/blog/handwriting-ocr) — Yes — modern OCR built on a vision-language model can read handwriting, because it reads whole words and fields in context rather than isolated characters. The honest caveat: handwriting accuracy is always lower than printed text, so the right way to use it is with per-field confidence scores and human review on the fields the model is unsure about — not as a fully hands-off pipeline.
- [How do you consolidate invoice data from dozens of vendors into one table?](https://inferio.xyz/blog/consolidate-invoice-data) — You consolidate invoice data by extracting every invoice — PDF, scan, or photo, from any vendor — into one shared schema: vendor, invoice number, date, line items, tax by rate, total. AI OCR does the normalization step that used to mean retyping each invoice into a spreadsheet, so the consolidated table builds itself as documents arrive instead of in a month-end crunch. Once the data is in one structure, spend reports, payables reconciliation, and duplicate detection stop being manual projects.
- [How do you turn paper documents into a searchable digital archive?](https://inferio.xyz/blog/paper-to-digital-archive) — Scanning alone doesn't digitize a document — it just produces a picture of it. A searchable archive requires one more step: extracting the content (vendor, date, amount, invoice number) into structured fields you can query, then using those fields to name, tag, and file each document automatically. That extraction step is what turns a folder of scan001.jpg files into an archive that answers questions.
- [Automating accounts payable for small teams](https://inferio.xyz/blog/automate-ap-for-sme) — For a small team, manual AP entry usually costs more in hours than in software — someone re-types vendor invoices into the accounting system by hand, one field at a time, several hours every week that never show up as a line item on a budget. An automated flow (upload → extract → review exceptions → export) turns that into a five-minute task per invoice batch, and the switch doesn't require a finance system migration to start.

## Links

- [Home](https://inferio.xyz)
- [Pricing](https://inferio.xyz/pricing)
- [OCR tool](https://inferio.xyz/tools/ocr-demo)
- [Developer documentation](https://docs.inferio.xyz)
- [Start with trial credits](https://merchant.inferio.xyz)
- [Frequently asked questions](https://inferio.xyz/faq)
- [Sitemap](https://inferio.xyz/sitemap.xml)
- Sales / partnerships: contact@lartech.xyz

---
Generated from the same source copy as the website. Canonical source: https://inferio.xyz