IDP

Read every document. Extract every field. Skip the manual entry.

Intelligent Document Processing for Australian businesses. We build LLM-powered pipelines that extract structured data from invoices, contracts, forms, and ID documents — with confidence scoring, human-in-the-loop review, and direct push to your existing systems. You own the code; we run the build.

96–99%

Field accuracy on invoices

−71%

Per-document time vs OCR

12 mo

Median ROI payback

3–4 wk

First pipeline live

Why IDP, why now

For thirty years "document automation" meant OCR plus rule-based templates — fragile pipelines that broke the moment a supplier changed their layout. The arrival of large multimodal models in 2023–24 changed the unit economics. A single LLM call can now extract structured fields from an arbitrary invoice layout for a fraction of a cent, with accuracy that previously required a six-figure platform contract.

According to Gartner's 2024 Hype Cycle for AI in CSPs, IDP has crossed the productivity plateau: 78% of mid-market deployments now realise measurable ROI within twelve months. The bottleneck is no longer the technology — it is finding a delivery partner who will ship a production pipeline you own outright, instead of a proof of concept that lives forever in a sandbox.

What we build

Invoice extraction. Supplier invoices, credit notes, statements. Extract supplier_name, ABN, invoice_number, invoice_date, line items, GST, total. Push directly to Xero, MYOB, QuickBooks, NetSuite.
Contract parsing. Customer contracts, employment agreements, NDAs, lease agreements. Extract parties, dates, key clauses, termination terms, monetary values. Index into your DMS or contract-management tool.
ID and KYC document processing. Driver licences, passports, Medicare cards, ABN searches. Validate against ASIC and ABR registries. Used by NDIS providers, financial services, recruitment.
Compliance forms. ATO, ASIC, Fair Work, NDIS service agreements, real-estate condition reports, vehicle inspection sheets. Structured output goes wherever your audit trail lives.
Custom document types. Anything with consistent semantic structure — medical referrals, insurance claims, supplier onboarding packs. We scope and price these in discovery.

How it works

Every IDP pipeline we ship has the same five-stage architecture:

1. Ingest

Documents arrive via email, S3 bucket, SharePoint, FTP, or direct API push. We deduplicate and route by document type using a small classifier.

2. Pre-process

Layout-aware OCR (AWS Textract, Google Document AI, or self-hosted PaddleOCR depending on data sovereignty needs). Image cleanup, deskewing, noise reduction.

3. Extract

LLM extraction call with the locked output schema. Few-shot examples for tricky fields. Returns structured JSON plus per-field confidence scores.

4. Validate

Deterministic rules: ABN check-digit, GST = 10% of subtotal, date sanity, totals reconcile to line items. Anything failing validation OR below the confidence floor is queued for human review.

5. Push + audit

Approved extractions push to downstream systems. Every pipeline run is logged: source document, extraction output, confidence scores, who reviewed, what they changed. Full audit trail in your database.

Engagement timeline

Week 1 — Discovery + sampling. Catalogue document types, pull representative sample of 30–100 documents per type, lock the extraction schema.
Weeks 2–3 — Pipeline build. Build extraction pipeline in staging. Working prototype runs against your real samples by end of week 2.
Week 4 — Accuracy tuning. Iterate prompts and add few-shot examples until each field hits its accuracy target. Documented weekly.
Weeks 5–6 — Integration + go-live. Connect to downstream systems. Configure human-review queue. Cut over from manual processing in production.
Months 2+ — Continuous improvement. Monthly accuracy review, prompt refinement based on human corrections, scale to additional document types.

Pricing

Single document type pipeline (one downstream integration)From $7,500 AUD

Multi-document workflow (3–5 document types, multiple integrations)$12,000 — $35,000 AUD

Ongoing infrastructure (LLM, OCR, hosting)$80 — $600 AUD/mo

Optional accuracy retainer (monthly review + tuning)$1,200 AUD/mo

All prices ex GST. Infrastructure costs passed through at cost. No per-document fees, ever.

Who this is for

IDP delivers the strongest ROI for businesses processing more than 200 structured documents per month manually: accounts payable teams, NDIS service providers, real-estate agencies, recruitment firms, professional services with high compliance volume, and any operation where a person spends >10 hours a week typing data from PDFs into another system.

It is a poor fit for very low-volume use cases (<50 documents per month) where the human time saved is less than the marginal infrastructure cost — though even at low volumes, IDP is often worth it for accuracy and audit reasons rather than time savings.

Frequently Asked Questions

What is Intelligent Document Processing (IDP)?

Intelligent Document Processing is the practice of using AI — typically a combination of OCR, layout-aware models, and large language models — to read, classify, extract, and validate information from documents that were previously processed by hand. According to Gartner's 2024 Hype Cycle for AI in CSPs, IDP delivers measurable ROI within 12 months for 78% of mid-market deployments. Unlike legacy OCR, IDP understands context: it can tell that "1,234.56" on an invoice is the GST-inclusive total even when the layout has never been seen before, because the surrounding words "Total inc GST" provide semantic context. We deliver IDP as production pipelines — not proofs of concept — that ingest documents, extract structured data, route to downstream systems, and surface only the edge cases that need human review.

How is IDP different from traditional OCR?

Traditional OCR converts pixels to text and stops there. You still need a human or a brittle template-matching script to figure out which numbers are dates, which are dollar amounts, and which are GST. IDP combines OCR with layout-aware vision models and LLMs to extract structured fields directly — "supplier name", "invoice date", "GST amount", "ABN" — regardless of the document's layout. According to Forrester's 2024 study on document automation, "AI-powered document processing reduces per-document handling time by 71% on average versus OCR-plus-templates approaches" (Forrester, The Total Economic Impact of Intelligent Document Processing, 2024). We build IDP pipelines that handle the long tail of layouts your suppliers send — not just the three standard formats your old template engine could parse.

Which document types can Iverel automate?

In production for Australian clients we currently process supplier invoices, purchase orders, customer contracts, employment agreements, ID documents (driver licences, passports), bank statements, utility bills, real-estate condition reports, vehicle inspection forms, medical referral letters, NDIS service agreements, and government compliance forms (ATO, ASIC, Fair Work). Anything with a consistent semantic structure — even if visual layouts vary — is a viable candidate. Highly free-form documents (handwritten notes, image-only forms with no structure) are harder but not impossible; we would scope these explicitly during discovery.

How accurate is LLM-based document extraction?

On structured documents like invoices and purchase orders we typically achieve 96–99% field-level accuracy after a two-week tuning cycle. On semi-structured documents like contracts and policy schedules, accuracy ranges from 88% to 95% depending on document complexity. The crucial design point: production IDP is not "set the AI loose and hope". Every pipeline we build includes confidence scoring per field, a configurable human-review threshold, and a feedback loop that improves accuracy as humans correct edge cases. We measure and report accuracy weekly. According to McKinsey's 2024 State of AI report, "organisations that pair generative AI with human-in-the-loop workflows realise 2.3× the value of organisations that deploy fully autonomous AI" (McKinsey & Company, The State of AI in Early 2024).

How long does an IDP project take to ship?

A single document type — for example, supplier invoice extraction — typically reaches production in 3–4 weeks. Multi-document workflows (invoices + purchase orders + contracts) take 6–10 weeks depending on integration complexity. We work in shippable increments: by the end of week one you will have a working extraction pipeline against ten of your real documents that you can review and validate. The bottleneck is rarely the AI — it is access to representative document samples, downstream system credentials, and the time of the person who will validate accuracy.

How does IDP integrate with our existing systems?

Every IDP pipeline we build emits structured JSON that can be pushed to any downstream system. We have production integrations with Xero, MYOB, QuickBooks, NetSuite, Salesforce, HubSpot, Zoho, monday.com, Airtable, Notion, Google Sheets, custom databases (PostgreSQL, MySQL, SQL Server), and any system with a REST or webhook API. We deliberately do not build proprietary connectors — we use the system's native API or a thin N8N adapter so you retain full control and visibility.

What does IDP cost in Australia?

A single-document-type IDP pipeline (one document class, one downstream system) starts at $7,500 AUD ex GST as a one-off project. Multi-document workflows typically range from $12,000 to $35,000 AUD depending on volume and integration complexity. Ongoing infrastructure costs (LLM API fees, hosting, OCR pre-processing) sit between $80 and $600 AUD per month for most clients, scaling with volume. We pass these through at cost. Replacement of legacy OCR licences (which often run $15,000+ per year) typically pays for the entire project within the first six months.

What about data sovereignty and privacy?

For sensitive documents (employment, medical, financial, legal) we offer pipelines built on Australian-region cloud (AWS Sydney, Azure Australia East) with all processing kept on-shore. We can run on locally hosted open-source models (Llama, Mistral) where regulatory requirements demand zero third-party API calls. By default we use Anthropic Claude or OpenAI GPT-4 with their Zero Data Retention agreements, which contractually prevent your documents being used for model training and delete them from provider systems within 30 days. We document every data flow in writing and review it with your privacy officer before go-live.

Why hire Iverel for IDP rather than buying an off-the-shelf platform?

Off-the-shelf IDP platforms (ABBYY, UiPath, Hyperscience, Rossum) charge per-document or per-seat fees that scale linearly with usage and lock you into proprietary data formats. We build IDP pipelines you own outright: code in your repository, infrastructure in your cloud account, no per-document fees, no vendor lock-in. We use the same battle-tested LLM and OCR primitives the platforms use — the difference is you keep the leverage. For high-volume use cases (>50,000 documents per month) the unit economics of owning your pipeline crush platform pricing.

See what IDP can extract from your documents

Send us five real documents from your most painful manual workflow. Within three business days we'll return a working extraction against your data and a fixed-price scope. No deck. No demo theatre. Just a working pipeline you can evaluate.

Send Your Documents →