Hospital Unit Patient Population Census Predictor

The Data

Patient population management data from thousands of hospitals describing things like the number of patients in units like the ICU, ER, etc. on a given time or day.

The Insight Hypothesis

Hospitals could be made more efficient if nurses and others involved in planning for new patient arrivals could know ahead of time how many patients would be present in their units at various forecast horizons into the future.

The Context

The company was new to integrating AI backed functionality into products, though an internal team of bright junior data scientists involved in largely development and devops activities was present. The company wanted a seasoned AI hand like Xyonix to work from a player-coach vantage to help accelerate the project.

The Solution

Xyonix began with some high level meetings to dive deep in the problem domain and to understand myriad product goals. Within 1 week Xyonix had functioning forecasting models and a plan to evolve forecasting capabilities over time. Within 1 month, naive (to serve as a comparative baseline) and classical machine learning baseline models were in place, exhaustive time series efficacy report cards were regularly generated, and a culture of experimentation and results driven discussions was initiated. Within 3 months, more sophisticated deep learning models were surpassing classical ML techniques. Around the same time, full production path dev and devops integration with company infrastructure was well underway.

The Impact

Xyonix was able to greatly accelerate a production path, rapidly train up internal data scientists, and initiate and evolve an AI company culture from project and product management through to executive level leadership.