💡 A Rostering Adventure Sparked by Curiosity

The funny thing is, the Emergency Department's Patient Service Associate rostering problem wasn't even something I experienced in my own day-to-day work.

But every month, I watched my frontline administrative colleagues wrestle with spreadsheet after spreadsheet and manual reconciliations. It looked exhausting. And being a naturally kaypoh person, I couldn't help wondering: Surely there had to be a better way.
 

What started as curiosity quickly snowballed into something much bigger.
 

Roster planning sounds straightforward on paper. In reality, it is a surprisingly complex operation involving multiple spreadsheets, emails, approvals, manual checks, and claims validation.

For years, that was simply how rostering worked and was accepted as just another part of the job. The process worked, but it was heavily dependent on manual effort. The more I explored the rostering challenge, the more I found myself thinking:
1.    The technology already existed.
2.    The process was fragmented.
3.    We had very little budget, but plenty of determination.

The question wasn't whether we could buy a solution but whether we could connect the dots using the tools and resources we already had. And that's when I accidentally wandered into the world of digital transformation.
 

The irony is that I am not a programmer. I'm not a data scientist. Nobody would ever mistake me for an AI engineer. Technology has never been my comfort zone. 

My usual approach to solving problems involved discussions, process mapping, and perhaps an Excel formula complicated enough to impress my colleagues for five minutes. Fortunately, I didn't have to figure everything out alone. My strategy was less "technology expert" and more "professional borrower".

I borrowed inspiration from Assistant Chief Medical Informatics Officer Dr Carmen Goh's work in digitalising Medical Officer leave processes and borrowed expertise from friends in Digital Group Office and Digital Think Tank  whenever things got complicated. Together, we eventually broke the problem into three manageable phases.
 

The first lesson I learnt was that AI and automation are only as good as the information they are given. Before we could automate anything, we had to clean up the process itself. That made Phase 1 the most important stage of the project. We transformed a largely manual workflow into a digital process and established a reliable data foundation. It wasn't the most glamorous work, but it was the most critical. Once that digital backbone was in place, Phase 2 focused on automating roster generation. Phase 3 tackled claims reconciliation. Each phase built upon the previous one. And each phase taught me something new. 

What surprised me most wasn't the technology. It was how much adaptability mattered. We encountered our fair share of roadblocks. Some ideas looked fantastic in meetings but fell apart during testing. Some solutions worked perfectly until they met real-life users and real-life workflows. More than once, our "final version" turned out to be nowhere near final.