Providus Healthcare is a US-based hospital chain. Its bed, staffing, and supply decisions were based on historical averages of patient demand, which did not account for seasonal surges or the differences between facilities, so resources at each site frequently fell short of what was actually needed. To address this, Bacancy built a healthcare demand forecasting model that combined patient and workforce data in Snowflake and ran on Databricks. The forecasts were served under role-based access and masking so protected data stayed governed across every facility. This improved forecast accuracy by 61%, reduced staffing spend by 47%, and replaced monthly averages with daily forecasting across every facility.
HealthCare demand forecasting deployed across every facility in the hospital chain
Patient and workforce data migrated into Snowflake with end-to-end validation
Automated daily forecasting built on Databricks, with retraining through MLflow
All patient data secured with role-based access and masking, matched to existing policy
Patient visits, admissions, and appointments were recorded in the EHR, and staff schedules and hours in the workforce system. No source brought the two together, so there was no combined record to build a forecast on.
Capacity and staffing were planned using historical averages. But these averages did not account for seasonal changes or the differences between facilities, so the beds, staff, and supplies set for a site often did not match the number of patients who actually arrived.
Forecasts had to run separately for every facility and for each unit and service line within it, resulting in thousands of series across the chain. The existing reporting tools could not handle that volume, so teams were left planning at a broad level without the detail they needed.
The data feeding the forecasts included protected health information, so any wrong exposure once outputs reached dashboards and reports carried real compliance risk. Each user needed the right level of access, and the same rules had to apply across every facility.
Bacancy’s healthcare data analytics team built pipelines that brought patient and workforce data into Snowflake as one source. Visit, admission, and appointment data from Epic was joined with staff schedules and hours, giving the hospital chain a single history to forecast from.
Our Databricks developers replaced the historical averages with ML models that forecast patient demand. Built with XGBoost, the models learned from patterns across the week, holidays, and seasonal surges, along with the recent history at each facility, so the beds, staff, and supplies planned for a site matched real demand.
We ran the model training on Databricks with Spark, which split the work across the cluster and trained the series in parallel. The thousands of forecasts the old tools could not handle were now produced on a daily schedule, with MLflow managing the models and moving approved ones into production.
Bacancy’s Snowflake developers served the forecasts through Snowflake and built the access model around the protected health information in the data. Row access policies, dynamic data masking, and role-based access were implemented to control what each user could see, and the same rules held across every facility.
Automated ingestion of Epic and workforce data into Snowflake
Dedicated demand forecasting for every facility and service line
Distributed daily model training and automated forecasting on Databricks
Role-based access and masking for protected health data
04
October 2025 – May 2026
61% improvement in healthcare demand forecasting accuracy over the historical averages
Overtime and agency staffing spend cut by 47%, as staffing matched demand and fewer shifts needed emergency cover
Bed and supply levels matched to real demand at each facility, instead of estimated in bulk
Each facility and unit moved to a daily demand forecast, replacing monthly manual averaging
Patient data stayed compliant across all facilities, with access governed by role
Manual healthcare demand forecasting replaced by an automated daily pipeline
| Source systems | Epic (Clarity, Caboodle), UKG |
| Data ingestion | Fivetran (change data capture) |
| Cloud data warehouse | Snowflake |
| Data transformation | Databricks |
| Forecasting | XGBoost |
| Model management / MLOps | MLflow |
| Orchestration | Lakeflow Jobs |
| Integration | Snowflake Connector for Spark |
| BI/reporting | Power BI |
| Scripting & automation | Python |
| Security & governance | Snowflake RBACdynamic data maskingrow access policies |
| Version control & CI/CD | GitGitHub Actions |
Get access to an experienced team of developers and engineers from Bacancy, handpicked to ace your goals. Kickstart within 48 hours, no-risk trial.
Years of Business Experience
Happy Customers
Countries with Happy Customers
Agile Enabled Employees