120x Faster Order Intake for a National Diagnostics Lab

The Challenge

A national diagnostics laboratory processing 45,000 specimens daily was drowning in manual data entry. Every incoming order — whether arriving by fax, PDF, or portal submission — required a technician to manually read the requisition, interpret handwriting, look up test codes in the catalog, and type the data into their LIMS. The process took an average of 6.3 minutes per specimen.

The downstream effects were severe. An 8% results delivery failure rate meant clinicians weren't receiving critical patient data on time. A 14.2% billing denial rate was hemorrhaging revenue through coding errors and incomplete insurance information. Staff were burned out from repetitive data entry across disconnected systems, and turnover in the order entry department was the highest in the organization.

The laboratory had tried incremental improvements — additional training, standardized forms, extra QA checks — but the fundamental problem remained: humans were manually transcribing data between systems, and errors were inevitable at scale.

Our Approach

DxLogic built an AI-powered document processing pipeline designed to handle the full spectrum of incoming order formats. The system uses computer vision and natural language processing to read incoming orders regardless of format — handwritten faxes, typed PDFs, and structured portal submissions are all processed through the same pipeline.

The AI extracts structured data from each order, validates it against the laboratory's test catalog, checks for common coding errors, and auto-populates the LIMS with verified data. Edge cases and low-confidence extractions are flagged for human review rather than auto-submitted, ensuring accuracy while still eliminating the majority of manual work.

Beyond order entry, DxLogic added automated results delivery — routing completed results to the correct clinician portal, fax number, or EHR integration — and upstream billing validation that catches denial-prone claims before they're submitted. The entire system was deployed in phases over 8 weeks, with the order entry automation going live first and the billing module following two weeks later.

The Results

Order entry time dropped from 6.3 minutes to 3.6 minutes per specimen — a 43% reduction. But the real story was in the downstream metrics: results delivery failures fell from 8% to 2.1%, meaning clinicians were getting patient data when they needed it. The billing validation module caught errors before submission, recovering $1.2M in annual revenue that had previously been lost to denials. Data entry errors dropped by 76%, and staff who had been spending their days on manual transcription were redeployed to higher-value work.