Insurers, hospitals, banks, ministries, law firms and energy operators all process documents they are not permitted to send anywhere. DocxIntel installs inside your own infrastructure with the models in the bundle, so the hardest documents in your sector are read on your hardware and nowhere else.
Accuracy is measured field-by-field against a human-adjudicated ground truth, not character-by-character. The document mix and per-field breakdown are published in full on the accuracy page, and your own threshold is agreed in writing during the Proof of Value.
The platform does not change between sectors. The document mix, the field schema and the downstream systems do — so each page below is written around what that industry actually receives every day.
Motor and medical claim packets arrive as one long scan of everything the claimant had. DocxIntel splits the packet, reconciles the claim against the policy schedule and flags duplicate or altered invoices.
Handwritten prescriptions, lab reports and discharge summaries are read inside the hospital network. Protected health information never reaches an external API, which is what the data-localisation rule requires.
Emirates ID, passports, trade licences, letters of credit and salary certificates are extracted, checked for completeness and examined for discrepancies without a page leaving the bank.
Handwritten Arabic application forms, permits, court files and decades of scanned archive run on sovereign infrastructure — including networks with no route to the internet at all.
Clause extraction, obligation tracking and privilege-aware redaction across contracts, filings and data rooms — with no third-party API in the chain of custody.
Bills of lading, delivery notes photographed on a phone at a gate, inspection reports and contractor invoices — reconciled from images that other tools reject outright.
Ask a bank CISO, a hospital CIO and a ministry records director what blocks document automation and you get one answer in three accents: the documents are not allowed to leave. Accuracy is negotiable; custody is not.
This is why on-premise is the default here rather than an enterprise upgrade path. The deployment model is the product decision.
These are the frameworks that come up in every evaluation across the six sectors. DocxIntel is infrastructure you operate, so your compliance position stays yours to certify — the architecture is designed to make that certification straightforward rather than to claim it on your behalf.
Central Bank expectations on outsourcing and where regulated data sits make a foreign processing endpoint an approval problem. A deployment inside your own data centre removes the outsourcing question rather than answering it.
Personal data in claim forms, KYC packs and citizen applications carries lawful-basis, minimisation and cross-border transfer obligations. Processing that never crosses a border does not need a transfer mechanism.
Health data generated in the UAE is subject to a data-localisation requirement. Reading a prescription inside the hospital network is the only version of prescription automation that satisfies it cleanly.
Financial-crime rules require records to be retrievable and evidenced for years. Field-level traceability means an extracted value can be shown back to the pixels it came from, on the page it came from.
Public-sector and defence-adjacent networks are frequently isolated by policy, not preference. An install that works identically with no route to the internet is a prerequisite, not a feature request.
Groups with European entities or European counterparties inherit GDPR obligations on top of local rules. Keeping processing in one controlled environment collapses several of those obligations into one.
A side-by-side of the document mix and the binding constraint per industry, as they come up in real evaluations.
Regulatory references are provided as context for evaluation, not as legal advice. Confirm the current obligations that apply to your entity with your own compliance function.
Cloud IDP platforms are capable products. The difference is structural rather than qualitative — per-page metering requires every page to reach the vendor, and self-hosting sits on the top tier.
Where documents are read
True air-gapped operation
Residency evidence for a regulator
Cost of 1M pages a month
Reprocessing the archive after a model update
Model weights
Self-hosting available on
Arabic as a primary script
Cost figures are our own arithmetic on publicly published list rates as of 2026 — roughly $0.0125 per page for agentic cloud parsing and around $0.30 per page on a usage-based IDP growth plan — not vendor quotations. Self-hosting availability reflects publicly documented plan gating. Verify against current vendor documentation before making a decision.
The path is the same in every industry. What differs is the field schema and which of your systems receives the output.
A representative sample matters more than a large one: the photographed delivery note, the handwritten Arabic form, the stamped invoice, the fourth-generation fax. If the difficult 5% is missing from the sample, the accuracy number is not meaningful.
The Proof of Value fixes the fee, the field-level accuracy threshold, the document set it will be measured on and the conversion price if it passes. Nothing about the commercial outcome is left to a later conversation.
Containers and open-weight models are installed on your hardware — air-gapped, private cloud or managed single-tenant. Your infrastructure team keeps root, the keys and the change window.
Document types, fields, validation rules and routing are configured on your actual documents. No template library is required, and no sample leaves the environment it was loaded into.
Results are adjudicated field-by-field with your reviewers, and the report shows per-field accuracy, confidence distribution and the exceptions that need a human. Miss the threshold and you keep the report and owe nothing further.
Volume grows without the bill growing. Adding a new document type or reprocessing the back catalogue is a configuration change and some GPU time, not a new line item.
The objections that decide whether document automation gets approved at all, answered without hedging.
No. It is one platform with one deployment bundle and one licence. What changes per industry is the document mix, the field schema and the systems output is written back into. Insurance claim packets, hospital prescriptions and trade finance files all move through the same six capabilities — Analyse, Identify, Classify, Map, Modify and Ask.
Thirty to forty-five days, fixed fee, deployed in your environment on your sector's real documents. The accuracy threshold and the conversion price are both agreed in writing before we begin.
