Balancing data and judgment in HR decision making
SteveA few weeks ago I did an HR Happy Hour Show with Joshua Gans, co-author of the excellent book Prediction Machines. On the show, we talked about one of the central ideas in the book - the continuing importance of human judgment in decision making, even in an environment where advances in AI technology make predictions (essentially options) more available, numerous, and inexpensive.
I won't go back through all the reasoning behind this conclusion, I encourage you to listen to the podcast and/or read the book for that, but I did want to point out another excellent example of how this AI and prediction combined with human judgment idea plays out in human capital management planning and decisions. A recent piece in HBR titled Research: When Retail Workers Have Stable Schedules, Sales and Productivity Go Up shares some really interesting findings about a study that aimed to find out if giving retail workers more schedule certainty and clarity would impact business results, and if so, how?
Some back story on the idea behind the study first. As demand planning and workforce scheduling software has developed over the years, and become much more sophisticated, many retailers now have the information and ability to set and adjust worker schedules much more dynamically, and almost in real time, than they had in the past. Combining sales and store traffic estimates with workforce planning and scheduling tools that are able to match staffing levels to this demand - store managers are, for the most part, able to optimize staffing, (and therefore control labor costs), much more precisely.
But while optimizing the staffing levels in a retail store sounds like a sound business practice, and makes the owners of the store happy (typically via reduced labor costs), it also often make the staff unhappy. In a software and AI driven staffing model, workers can find their schedules uncertain, changing from week to week, and even find themselves losing expected shifts on very short notice, sometimes less than two hours.
The data and the AI might be 'right' when they recommend a set of staff schedules based on all the available information, but, as we will see in the research referenced in the HBR piece, the data and the AI usually fail to see and understand the impact this kind of scheduling has on the actual people that have to do the actual work.
You really should read the whole piece on HBR, but I want to share the money quote here - what the researchers found or recommended would be the best way for a retailer to incorporate these kinds of advanced AI tools to help set retail store worker schedules:
At the start of the study, we often heard HQ fault store managers for “emotional scheduling” — a script pushed by the purveyors of scheduling software. “In measuring customer experience and making decisions related to a labor model, retailers should rely solely on facts. Too often, changes are made because of an anecdotal or emotional response from the field,” notes a best practices guide from Kronos.
However, our experiment shows that a hybrid approach of combining algorithms with manager intuition can lead to better staffing decisions. While our experiment provided guidelines for managers, it still allowed the managers to make the final decision on how much of the interventions to implement. The increase in sales and productivity witnessed at the Gap shows that retailers stand to benefit when they allow discretion to store managers.
What were some of the benefits of giving managers at least some discretion over scheduling, even when the AI made different recommendations?
When managers could give more workers more 'certain' or predictable schedules, most of them benefited from ability to predict commute times, ability to schedule things like education, child care, other jobs, and enabled them to connect more deeply with customers and co-workers. In short, they were all happier, and this tended to lead to better work performance, better customer service, and in the case of the stores studied by HBR - increased revenues and profits.
In time, maybe the AI will learn to understand this, this nuanced, subtle, but important impact that work schedules have on workers, and how that impacts business results. But until then, it seems like it's best to let the AI make recommendations on the optimal staffing decisions, and let the managers make the final call, based on what they know about their staff, their customers, and well, human nature in general.
Have a great day!
