Sometimes great ideas originate from a disagreement with another great idea. This was the case less than a year ago when Musti Group’s HR Director contacted us needing help. It turned out, the HR department and the business managers had disagreements regarding workforce planning and certain related practices.
As a side outcome of our people analytics and workforce optimization project, the HR department ended up having more accurate sales forecast than the business had.
Questions We Needed to Answer
Early on, we were curious about answering questions like:
“How likely were the best performing sales people allocated to the best sales times?”
“How often are the days/times, when a store is over/under staffed?”
“How could we better benefit from past performance data and build more accurate predictions to optimize operations and workforce management?”
Arguments had been going back and forth. They were based mostly on intuition and gut feeling without data and analysis to back their point of view. To disagree well, we had to understand the problem better by conducting further analysis.
Embarking On An Iterative Journey
Peoplegeeks took this practical opportunity to sharpen thinking around the workforce practices of the Nordic Retailer Musti Group. To begin with, we started to gather our understanding of the situation. We listened carefully to different points of view from the key stakeholders, looking at the data and building comprehension of the bigger picture. In other words, we collected evidence to reason the best actions. And yes, we had to tolerate the risk that we wouldn’t necessary be right with our initial hypothesis.
This project was built in an agile manner. We gradually built our comprehension of the right problem, discovering new valuable aspects to work on, validating possible solutions and defining next steps as our mutual understanding crystallized. The progress has been a constant interplay between the business needs, people practices and the opportunities provided by data and advanced analytics.
Examples of Our Findings
Along our people analytics journey we have made several discoveries by analyzing the data and discussing with different stakeholders. Some of the insights first seemed counter intuitive and surprising. Others were more expected and supported our initial hypothesis.
As an example, on a high-level investigation, the data suggested that the 0-hour contracted employees performed the best in terms of sales per hour. However, as we continued our analysis with more detailed level data, we found proof that this group of employees weren’t necessarily the best performing group after all.
It turned out that 0-hour contracted people were more likely to have worked their shifts in high-demand rush hours. As we compared average shopping basket sizes of store personnel, with the help of receipt data, the more experienced personnel tended to outperform the more inexperienced group. Data also revealed that the most skilled personnel were selling a broader share of the overall store selection of products. This was a key to further prove that experience and product knowledge matters, yet the workforce practices at the time weren’t benefitting from these findings, leading to sub-optimal outcomes.
From Discoveries to Actions
As with all analytics projects, their value can only be realized after we take action and impact people’s behavior. In our case, we did not necessarily need to make any ground-breaking discoveries. Rather, the value was found in a series of smaller insights that validated the interdependence of workforce management practices and their consequent business and people outcomes.
The actions we have taken fell under three topics:
1. Enabling workforce fluctuation needs:
– Optimized contract types, contract hours and headcount for greater efficiency
– Calculated a new demand-based leveling periods, enabling higher contract hours for employees by making it easier to schedule hours for varying demands through the year.
2. Helping store managers focus on productive work:
– Reduced participation time of hiring managers from 16h to 4,5h and improved the quality of new hires
– Minimized the time required from store managers for scheduling shifts with centralized scheduling and workforce planning practices
3. Increasing scheduling efficiency and impact:
– Redesigned workforce scheduling and planning activities to enable automation and optimization
– Optimized staffing hours to match expected traffic utilizing machine learning models
– Identified top performing teams and individuals and targeting them to most important shopping periods on daily/weekly basis.
Actions in the first two categories gave us the baseline and aimed to make the processes more efficient and scaleable. The third one aimed at taking advantage of the series of discoveries we had made along our journey and optimizing the outcomes by using analytical models to automate workforce scheduling with the goal “right people in the right place at the right time”.
How Things Have Changed
Previously each store manager had overseen workforce planning and shift scheduling for their own store. However, they lacked means to fully benefit from the vast available data and therefore they mostly relied on their intuition. Also, their actions were severely limited due to having only a small headcount for one individual store that did not offer much opportunity for variation. On the top, what was perceived as “optimal” action from an individual store managers perspective, was anything but optimal from the overall perspective, leading to inefficiencies and a waste of resources. This lead to sub-optimal outcomes and constant fire-fighting mode to cope with unpredictable changes.
We tackled this challenge with advanced analytics to harness the data for identifying optimal practices. We accurately predicted workforce demand variations and designed centralized workforce planning and scheduling practices serving multiple locations on an regional level.
Response to Workforce Demand Variations
As opposed to reporting, where all the facts must be right, analytics and predictive models work with odds. The goal is simply to be more right, more often. Therefore, it was enough to simply increase the efficiency and impact, even just marginally compared to the past, in order to come out on the winning side. Along the journey we were able to benefit from more data, combine various data sets and tailor the optimization models, we could increase our odds even further.
The Game Changing Moment
To develop the solution for optimized scheduling, we had to create accurate demand model. A big part of the demand for store personnel is linked to customer traffic and sales. The business executives were only able to provide us budgets that were on a daily level. For our scheduling purposes, we needed to have hourly-level predictions for each store for every day of the year.
For this purpose, we built a model using machine learning and vast amounts of historical data. As we were validating the outcomes of the machine learning model against actual and budgeted sales – our model proved to be more accurate than the budgets made by business!
- In the beginning, the complexity of advanced analytics may appear overwhelming. There are huge amount of opportunities and challenges to tackle, variables to control and data challenges to overcome. However, as soon as you prioritize the potential use cases and start, the learning curve tends to be steep.
- Real advanced analytics capabilities can be hard to find these days. Companies should find the right balance for doing projects in-house versus outsourcing expertise. Especially in the beginning, companies are likely getting ahead a lot quicker by outsourcing the required skills for experienced experts and building internal capabilities along the way.
- Business executives often don’t care how the analytics are conducted and what solutions are used, as long as they get desired results. Professional consultancies focus on building scalable analytical solutions, whereas individual experts may make themselves indispensable.