AI-Assisted Shift Scheduling and Funding Exports for a Japanese Care Provider
Shifts had to balance staff, volunteers, qualifications and patient need, which is too complex for a spreadsheet. We built AI-proposed scheduling with a human always deciding, and generated government funding exports from the same record.
- Service
- AI Business Applications
- Industry
- Healthcare & Social Care
- Region
- Japan
- Engagement
- Lab-type ODC (dedicated team)
Outcomes
- AI-proposed, human-approved rosters
- Planned vs actual in one record
- Funding claims generated automatically
Stack
Python · Scheduling optimisation · Web app · PostgreSQL · Cloud
Written byOriental Nippon Team
A Japanese care provider delivers care both on its own premises and in the community. Its workforce mixes employed support staff and volunteers, each with different availability, qualifications and limits. Its funding comes through a government scheme that wants activity reported in a specific format.
All of that was coordinated by hand, and then translated by hand into the funding claim.
The problem
- Too many constraints for a spreadsheet. Staff, volunteers, qualifications and patient needs all had to line up.
- Two kinds of care, one picture. On-site and off-site visits were scheduled differently but had to be seen together.
- Plan and reality disagreed. Activity was recorded separately from the schedule.
- Claims rebuilt from scratch. The funding submission was assembled from several sources.
- High stakes. A scheduling mistake could mean a care visit not happening.
| Before | After | |
|---|---|---|
| Scheduling | Manual coordination | AI-suggested, human-approved |
| Planned vs actual | Two separate records | One record |
| Funding export | Rebuilt by hand | Generated in the scheme's format |
| Validation | After submission | Tested against real submissions before go-live |
What we built
AI-supported scheduling
Proposes shift assignments against availability, qualifications and location. A person always makes the final call.
Activity tracking tied to the schedule
Completing a visit updates the schedule itself, so planned and actual are one record.
Compliant funding exports
Generated from the activity record in the format the national funding system expects.
Each shift moves from suggestion to funding claim on one record:
Technology and the role it played
| Technology | Role in this project |
|---|---|
| Python | Scheduling logic and constraint handling |
| Scheduling optimisation | Proposes assignments against availability, qualifications and location |
| Web application | Coordinator view for review and approval |
| PostgreSQL | Staff, volunteers, patients, sites and activities |
| Cloud hosting | Access from on-site and community care |
How we delivered it
Our ODC team mapped the full constraint set first: who can do what, where, when, and under which rules. Scheduling suggestions only came after that model existed.
- Map the constraints. People, qualifications, sites and funding rules.
- Foundation. Data model for staff, volunteers, patients, sites and activities.
- Scheduling, then the assistant. Suggestions always reviewable before they take effect.
- Activity tracking. Completion updates the schedule, not a parallel log.
- Funding export. Validated against real past submissions.
- Pilot. One team through a full funding period.
01Discovery
2–3 wks
Constraint set, funding requirements, care patterns
02Foundation
2–3 wks
Data model, environments, access control
03Scheduling and assignment
4–6 wks
Rosters, constraints, assistive suggestions
04Activity tracking
3–4 wks
On-site and off-site capture tied to the schedule
05Billing export
2–3 wks
Format compliance, validation against real submissions
06Pilot period
3–4 wks
One team through a full funding cycle
Risks and how we handled them
| Risk | How we handled it |
|---|---|
| A missed care visit | Every suggestion is reviewed by a coordinator before it takes effect |
| Funding claims rejected | Exports validated against real past submissions |
| Plan and reality disagreeing | Activity recorded against the schedule itself |
Outcome
Shift management became one activity instead of a coordination problem solved by hand every week. The funding submission now comes from the same record that runs the care.
Connecting to the national funding scheme became straightforward and cheap. That was the outcome the client cared about most, and the one we tested most carefully before go-live.
What the client keeps
- The production system and its source, in their own repository
- The constraint and qualification model
- The export specification and its validation cases
- Operational runbooks
Sources and further reading
- People + AI Guidebook, Google PAIR
- Guidelines for human-AI interaction, Microsoft HAX Toolkit
Facing something similar? This is the kind of work our AI, Data Science & Automation team delivers every day. Talk to us about your stack.
