Case study · 01

Workforce AI Scheduling

Predicting and fulfilling workforce needs for hospital systems.

Lead Product Manager · Aya Healthcare · 50-person cross-functional org · $5M operating budget

$40M

MSP revenue supported

$500K

SaaS revenue, enterprise

50+

Person org

AI products shipped

Context

The Problem

Hospital staffing generates thousands of daily decisions — vacancies, sick calls, overtime, incentives, schedule optimization. Existing workforce management tools gave managers mountains of data. What they lacked was guidance on what to actually do with it.

The problem wasn't prediction. It was decision quality. Hospital leaders needed to act quickly and confidently in high‑stakes, high‑variability environments where the cost of a wrong call is measured in both dollars and patient care.

Product

The Solution

Workforce AI is a suite of AI-assisted tools that surface staffing recommendations directly inside the workflows hospital operations leaders already use. Rather than presenting raw forecast data and leaving interpretation to the manager, the system translates ML outputs into specific, actionable guidance — recommended schedule changes, predicted shortage windows, and optimization opportunities across shifts, units, and roles.

The core products: Optimized Schedule generates AI-recommended staffing plans. AutoBalance handles real-time shift adjustments. Lookahead flags emerging coverage gaps before they become operational problems.

Scope

My Role

I owned product strategy and delivery for the full Workforce AI initiative, working across executive leadership, workforce operations experts, engineering, design, implementation, and AI/ML teams.

Products shipped: Optimized Schedule · AutoBalance · Lookahead · Azure AI/ML forecasting architecture · Self‑service onboarding.

Approach

Key Decisions

Scalable onboarding before new feature development

The roadmap had strong pull toward shipping a new tool to recommend open shift changes to managers. The more pressing constraint was that onboarding new health systems required heavy manual support — which was becoming the ceiling on growth. Prioritizing self-service onboarding infrastructure first created a foundation that could scale without proportional headcount. A useful feature sitting behind a broken onboarding process doesn't compound. A scalable delivery foundation does.

Modernize the forecasting architecture

The existing forecasting engine was a decade-old time series system held together with patchy SQL work — functional but increasingly difficult to maintain, extend, or improve. Migrating to a Python-based application on Azure AI/ML replaced brittle legacy infrastructure with a modern, maintainable stack that the team could actually own and iterate on.

Restructure the team for ownership and quality

As the portfolio grew, two structural problems were limiting delivery. Ownership was diffuse — all teams were nominally responsible for everything, which meant accountability was effectively owned by no one. And the ratio of junior and outsourced engineers to senior US-based engineers was too high to sustain quality at pace. Reconfiguring around clear US-based senior engineering and PM ownership, with an improved seniority ratio, addressed both.

Reflection

What I Learned

The most durable wins came from reducing the distance between a decision and the data that supports it. Shipping AI features is the easy part; aligning operations leaders, engineering, and clients on what "good" looks like is the work.

Back to home Talk about this work →