Staff optimization has been one of the rising topics in revolutionizing brick-and-mortar. Read about how we at Pygmalios approach this task with the power of a digital doppelgänger.
In our previous article on Digital Twins of Retail Stores, we linked this popular concept to different types of in-store data we collect and analyze, such as these:
However, what we currently focus on the most regarding digital twins of retail stores is planning and optimization. This means improving processes that take place right inside the stores, such as staff optimization or employee task management.
Our colleague from the data science team Matus in his talk at the COMIT Digital Twins Awareness meetup gave a deeper look into how he and his team approached staff optimization.
You can watch the talk at the end of this post but if you’re busy (and that’s usually the case) though curious at the same time, just continue reading to get most of your questions answered.
Essentially, we want to create an employee schedule that looks something like this:
You’ve got a monthly schedule of when each employee should come to work and how long should his shifts be. Some days, though, the employee is off or on a vacation or seeing a doctor.
Taking the number of customers in the store and the number of transactions into account, does the employee schedule reflect the actual employee demand?
This is the challenge for most physical retailers and the one digital twins are very useful for.
To give you an idea of the whole process behind generating these schedules for staff optimization, have a look at the charted process below:
We start with historical customer traffic and occupancy data, so how many customers entered the store and how many were there at the same time. Additionally, we may add transactional data to look at different service times.
This data is used to predict the employee need for the next month. We predict an ideal headcount a month in advance and for every 15-minute slot for each day of the month.
We combine these predictions with:
The whole data model is then run through a constraint solver that produces the schedule that fits the need and constraints.
This is by no means the final look. The final schedule is adjusted and modified by the end-users, usually the store managers. It changes over time and as the month progresses.
Constraint optimization is the general problem area we work with by modeling the problem using:
Variables are something like states, for example, an employee either is or is not at work. Constraints provide rules for the variables, such as opening hours or required gaps between shifts. Objectives are the optimization goals we aspire for and try to get to as close as possible.
Once we have the data model, we push it through the solver whose job is to identify the feasible and ideally optimal solutions out of a very large set of candidates.
If you’re looking for a nice practical example using Google OR-Tools, you can check the video at the end of this article (starts at 15:38).
Yes, retailer feedback is a key to every successful workforce optimization task. Employee schedules constantly need to adapt to different retailers and unique stores.
Our approach is to create suggestions which are then iterated over with the retailer. We take in the feedback and the changes they made to improve our future schedule optimization and personalize it.
Data is just one part of the puzzle. The other big part is the rules and preferences made by retailers. We found that there is a trade-off between how flexible we should be towards changes originating from individual store managers and trying to find a common ground for all the stores.
Wrapping up, we at Pygmalios look at the concept of digital twins of retail from these two points:
Watch the talk: