Hospitals, clinics and payers hold the documents least permitted to travel and hardest to read: handwritten prescriptions, faxed lab reports, discharge summaries, pre-approval forms. DocxIntel reads them on your own hardware, with the models in the bundle and no external API in the path.
Accuracy is measured field-by-field against a human-adjudicated ground truth, not character-by-character. Medication, dosage and identifier fields are held to tighter confidence thresholds than descriptive text, and your own threshold is agreed in writing before the Proof of Value begins.
Hospitals digitised their systems long before they digitised their paper. Prescriptions are still handwritten, referrals still arrive by fax, and pre-approvals are still assembled by hand from four different documents.
Handwritten prescriptions are read into structured medication, dosage, frequency and duration fields, with the prescribing clinician and date captured alongside. Ambiguous entries are flagged, never guessed.
Analyte names, values, units and reference ranges are recovered from tabular lab reports, including faxed and multi-generation copies where the table rules have half disappeared.
Admission and discharge dates, diagnoses, procedures performed, length of stay and discharge medication are extracted from long-form narrative summaries.
Pre-authorisation forms are assembled and checked for completeness against the payer requirement before submission, so packs come back for clinical reasons rather than missing pages.
Referral letters, consent forms and intake paperwork are classified and indexed, including signature and checkbox state, so the chart is complete before the appointment starts.
Source documents for DRG and clinical coding are indexed and cited, and scattered paper is consolidated against the right patient identity across documents that spell the name three different ways.
Pre-authorisation is where clinical time is lost to administration. The pack is rejected for a missing page, resubmitted two days later, and the patient waits. Every step below runs inside your network.
Nothing in this sequence involves an external API call, which is what makes it viable under a data-localisation requirement rather than merely fast.
One department, one document type, measured on your own records before anything else is discussed.
Usually outpatient pharmacy, the pre-authorisation desk or medical records. We ask for the documents your staff find hardest — the illegible prescription, the fourth-generation faxed lab report — because those decide whether the accuracy number survives production.
Because processing happens inside your network, the review is about internal access control and audit logging rather than cross-border transfer. Data-flow documentation and deployment evidence are provided for your assessors before installation, not after.
Containers and open-weight models are deployed on your hardware in the segment your policy requires, air-gapped or private cloud. Your infrastructure team holds root and the encryption keys, and no patient record is ever visible to us.
Document types, field lists and validation rules are configured on your own forms. Medication, dosage and patient-identifier fields are set to stricter confidence thresholds so anything doubtful lands in a human queue by design.
Accuracy is adjudicated field-by-field by your pharmacists, coders or records staff. The report shows per-field accuracy, the confidence distribution and the review load you would actually staff. Miss the agreed threshold and you keep the report and owe nothing further.
Adding a second department, a second clinic or a decade of archived paper records is a configuration change and some GPU time. The licence does not move because the page count did.
An indicative schema. The final field list is agreed against your own forms and your own payer requirements during the Proof of Value.
Regulatory references on this page are provided as evaluation context, not legal advice. Confirm the obligations that apply to your facility with your own compliance and information-governance functions.
Cloud IDP platforms read medical forms well. The problem is structural: a metered service has to receive the page before it can read it, and the page is a patient record.
Where patient documents are read
Fits a health data-localisation requirement
Works on an isolated clinical network
Parsed PHI cached outside your control
Digitising a decade of paper records
Illustrative cost at 100,000 pages a month
Handwritten Arabic clinical notes
Self-hosting available on
Cost figures are our own arithmetic on publicly published list rates as of 2026, not vendor quotations. Caching behaviour and self-hosting availability reflect publicly documented behaviour of metered parsing services. Verify against current vendor documentation before making a decision.
The questions that decide whether clinical document automation is approvable at all, answered without hedging.
No. The models ship inside the deployment bundle and run on your own GPUs, so a prescription is read by software sitting in the same data centre as the patient record it belongs to. There is no external inference endpoint, no telemetry beacon and no fallback cloud call when confidence is low — an air-gapped install behaves identically to a connected one.
Thirty to forty-five days, fixed fee, deployed inside your clinical network on your own records. The accuracy threshold and the conversion price are both agreed in writing before we begin.
