As part of developing the Rosterlab AI staff scheduling and rostering platform, we've had to hone our rostering AI's accuracy. Thankfully we are fortunate to have the time and support of many excellent individuals who perform the rostering function in healthcare. We've learned what matters in an outstanding roster and the trade-offs that often have to be made.
By trade-offs, we mean the (sometimes implicit) decisions a rosterer has to make that satisfy one criteria at the expense of another, such as giving a shift to someone that would rather not work so we don't understaff that shift. Creating a good roster is akin to solving a sudoku, placing shifts on a grid to fulfil as many criteria as possible.
Unlike sudoku, though, there is no end condition with rostering, which leads to our first significant trade-off. How much time can a person afford to spend making, improving, toiling, and tinkering at a roster? Especially when there is no certainty that the improvements would lead to happier workers or a better skills-mix each day. Removing a single shift from a roster can require a cascade of changes. Making sure you have enough staff working on each day can create new gaps in skills mix or fairness or contractual hours, or individual requests. The Rosterlab AI can evaluate thousands of potential rosters a second, its artificial brain exploring all the rosters that aren’t improvements to find one that is.
The individual requests bring us to another critical time-related constraint. When does the rosterer start working on the following roster, and when do they stop? Suppose they start building their roster too early. The individual requests may change too late, and there is not enough notice of the following schedule.
Staffing demands can also be incredibly complex when creating rosters. There is an ideal ratio of differently qualified employees required for any given shift that is often challenging to achieve with the available staffing pool. This is even before considering the fluctuations in available hours caused by leave. The complexity of just fitting in the right shifts is further compounded by contractual obligations, union rules and best practices obligations that limit what shifts can be given when and to whom. It is not uncommon to have flexible interpretations of some of these restrictions if the alternative is to have an understaffed day. And when presented with the choice of having an understaffed day or ignoring an individual preference or request, the request often doesn't stand a chance.
A roster can fulfil the needs for staffing coverage and contractual obligations and still be a flawed roster. Each individual working in the roster would have requests and preferences that would ideally be met if possible. Realistically the rosterer can only meet a subset of these requests, and some are easier to grant than others. The trade-off is not only time spent refining a roster to meet more of these individual requests but also fairness in how many preferences are met. Some requests already line up with staffing needs, and some can be coincidentally fulfilled. One person may get most of their request fulfilled by pure random chance, while others get few. This imbalance invites accusations of bias and preferential treatment that can reduce overall staff happiness. All these trade-offs can discourage a person from trying to make the best roster.
Good enough is sometimes the only achievable standard with the time burden and complexity of rostering. Good enough is also more readily reached with common rostering practices such as self-rostering or rolling rosters.
Another common practice is to repeat a previously working roster and only make the necessary swaps and changes. These all lack the flexibility to discover a schedule that optimises both workplace and staff satisfaction.
If you'd like to explore the possibility of better than good enough, of better than humanly possible, contact Rosterlab about AI rostering and staff scheduling and its benefits to your organisation.