The workflow before#
Elara Caring runs home-based care for more than 65,000 patients a day across 225 locations in 16 states, built through acquisitions that each brought their own EMR and EHR systems. Intake staff read referral packets by hand, keyed patient data into separate systems, verified insurance one case at a time and priced bids from spreadsheets. A single referral took 4 to 6 hours of staff time, and it took 3 to 5 days to move a patient from referral to admission. Manual entry held data accuracy at 85 to 90%, and slow bid responses cost Elara cases it should have won.
What we built#
We built an automated intake pipeline on Azure. Referral documents land through Azure Data Factory, and Azure Databricks runs OCR to pull text from scanned pages, then NLP models classify each document and extract patient details, diagnoses and treatment plans. Automated validation checks every record and routes only the exceptions to an intake specialist.
On the same data we built underwriting and bidding models: risk scoring for patient complexity and a profitability model that estimates the margin on each case before Elara bids. Insurance eligibility checks run automatically, so a bid can go out the same day the referral arrives.
Clean records flow into Azure Synapse Analytics, and Power BI dashboards show every referral by stage, flag stalled cases and track margin by service line. The work ran inside a HIPAA-compliant environment designed with Elara's security team.
Before and after#
| Measure | Before | After |
|---|---|---|
| Referral to admission | 3 to 5 days | Under 24 hours |
| Staff time per referral | 4 to 6 hours | Under 45 minutes |
| Bid response time | 2 to 3 days | Same day |
| Intake data accuracy | 85 to 90% | 98% |
What's next#
The intake pipeline now feeds clean, structured patient data into one warehouse. The next step is extending the same workflow pattern to claims and denial management, where specialists still review most cases by hand.
