Background

Work shift planning in the public healthcare sector is largely a manual, recurring administrative burden. Shift planners spend significant time each planning cycle building rosters by hand, reconciling employee requests, absence rules, competence requirements, and labor agreements — often in spreadsheets or legacy scheduling tools that offer little intelligence beyond basic rule-checking. In Finland's wellbeing services counties (hyvinvointialueet), this work is repeated across dozens of units and thousands of employees, with limited standardization between them. At the same time, generative and predictive AI have matured enough to handle constraint-heavy optimization problems, interpret free-form employee requests, and flag risk patterns in structured data — capabilities that map directly onto the pain points of shift scheduling, but that have not yet been applied systematically in this domain.

Challenge

How can a simple, practical AI-powered shift scheduler be built for use in the public healthcare sector — one that takes over as much of the routine planning work as possible while keeping humans in control of the outcome? The project will investigate and test AI use across the following areas, among others: - Employee requests: capturing free-form preferences and constraints (e.g. "I need to be home by 16:00 on Tuesdays," "I want as many weekend shifts as possible") and translating them into a schedule the AI can optimize against. - Workload and burnout risk: tracking workload over longer periods (not just single shifts or weeks) and alerting management when an employee shows signs of being at risk of burnout. - Feedback loop: giving employees a channel to tell management how they feel about the resulting schedules and what should change, and feeding that signal back into future planning rounds. - Human-in-the-loop design: structuring the system so that the AI produces ready-made schedule proposals, and the only action left for humans is to approve or reject them — and testing how far that ambition can realistically be pushed.

Goal
1. Design and test an AI-based scheduling engine that turns employee requests and constraints into draft shift plans for a healthcare work unit.

2. Build a workload-tracking mechanism that monitors employees over longer periods and raises early alerts for burnout risk.

3. Create a feedback loop through which employees can report on schedule quality and fairness, and route that input back to management and into future scheduling rounds.

4. Determine how close the system can get to a fully AI-generated, human-approved-only workflow — and identify where human judgment is still required.

Scope
The project is not aimed at a commercializable or production-ready scheduling product, but at validating whether AI can meaningfully take over the core mechanics of shift scheduling in a public healthcare setting. The scope is limited to one or two representative work units (e.g. a ward or care team) with a simplified but realistic rule set, so that most effort goes into testing and evaluating the AI's scheduling, risk-detection, and feedback-handling capabilities rather than building a full-featured, organization-wide system.

Expected Outcomes
- A working prototype of an AI shift scheduler that generates draft rosters from employee requests and constraints for a pilot work unit.
- A workload-risk alerting mechanism based on workload tracked over longer periods.
- A functioning feedback loop between employees and management on schedule quality.
- Findings on how much of the scheduling process (proposal generation, risk detection, and iteration) can be handled by AI with humans only approving or rejecting outcomes, and where human oversight remains necessary.