Twelve thousand legacy paper charts moved into the EHR without anyone typing a record. Scans flow through OCR and an AI extraction layer that pulls demographics, allergies, and medication history into structured fields, while low-confidence pages route to a short human review queue.
Typical results
6 wks
Project duration
96%
Fields auto-extracted
4%
Pages to human review
Figures reflect typical results for this kind of automation.
The challenge
Forty years of practice lived in filing cabinets: 12,000 paper charts with handwritten notes, allergy lists, and medication histories the new EHR couldn't see. Manual data entry was quoted at 18 months and a small fortune, doctors kept walking to the archive mid-consultation, and a misfiled allergy was a real clinical risk.
Before and after
Migration timeline
−92%Staff minutes per chart
−90%Pages needing manual entry
−96%The automation
We set up a batch-scanning pipeline: charts go through a high-speed scanner, OCR converts every page, and an AI extraction layer reads the results — pulling demographics, allergies, diagnoses, and medication history into the EHR's structured fields with a confidence score per item. Anything below the confidence bar routes to a review queue where a medical secretary confirms or corrects it in a side-by-side view. Original scans attach to each record, so nothing is ever lost.
How it works
- 01
Batch scanning
Charts are pulled by the box and fed through a high-speed scanner, hundreds per day.
- 02
OCR every page
Each page is OCR'd, including typed forms and most handwritten sections.
- 03
AI extraction
The AI layer pulls demographics, allergies, diagnoses, and medications into structured EHR fields.
- 04
Confidence gate
Every extracted item carries a confidence score; low scores route to the review queue.
- 05
Human confirm
A secretary verifies flagged items side-by-side with the scan — about 4% of pages.
Starts from
- 12,000 paper charts
- Batch scanning
Delivers
- Structured EHR fields
- Human review queue
- Verified records
The impact
All 12,000 charts migrated in six weeks instead of the quoted 18 months, with 96% of fields extracted automatically and only 4% of pages needing human review. Doctors now see full allergy and medication history on screen, and the archive room became a second treatment room.
Questions about this automation
How does the system handle handwritten notes?
OCR reads typed forms reliably and most clear handwriting. Every extracted field gets a confidence score, and anything below the threshold, such as a faint allergy note or an unclear dose, goes to a review queue instead of the record. A staff member checks it against the original scan before it is saved.
Which data is extracted into the EHR?
The extraction step targets the fields clinicians need on screen: demographics, allergies, diagnoses and medication history. The practice decides which fields map to which EHR sections. The full scan of every chart is attached to the patient record, so anything not extracted as a field is still one click away.
How is patient data handled during the migration?
Scans and extracted data move only between the scanner, the extraction step and the practice's EHR, with access limited to the staff who work the review queue. The setup follows the practice's own privacy requirements. It is a multi-step build, usually rolled out in stages, starting with a small test batch of charts.