Have you seen the GE commercial about “Molly, the Kid Who Never Stops Inventing”? Every time I see it, I’m reminded about how robots are becoming a greater part of our workplace. And that’s not a bad thing, but it does mean that we need to get more comfortable with today’s technology concepts.
That’s why I’m very excited to share today’s interview with you. A few weeks ago, I wrote about Kronos’ new next generation workforce management solution called Workforce Dimensions. Dr. Thomas Walsh is the director of data science at Kronos, where he leads the application of cutting edge machine learning (ML) and artificial intelligence (AI) techniques to workforce data. His work includes designing AI solutions around compliance and business forecasting for Workforce Dimensions.
Walsh received his PhD in computer science from Rutgers University and has held research positions in both academia and industry. His work has covered diverse fields from robotics to educational modeling, and now workforce management. He is the author of over 30 peer-reviewed publications in the fields of ML and AI. And he very graciously agreed to share a little of his expertise with us about machine learning.
Dr. Walsh, can you briefly describe machine learning and how it differs from artificial intelligence?
[Walsh] Machine learning (ML) and artificial intelligence (AI) can be tough fields to pin down.
Artificial intelligencerefers to a collection of subfields thatsolve complex problems associated with human intelligence and/or interacting with the world. These subfields include computer vision, natural language processing, robotics, and machine learning.
Machine learninggenerally covers methods that build models of complex data.
There are many types of ML approaches, and it is important to choose the right type of solution for the specific business problem at hand. For instance, in the workforce management space, Kronos uses unsupervised ML algorithms that uncover patterns in data to find potential compliance violations. But in the Workforce Dimensions product, Kronos uses a different type of ML, supervised ML regression, for business volume forecasting.
Each of these types of ML has its own idiosyncrasies, but the one thing they all have in common is that they are driven by the data they are given. In the same way a new recruit might need time on the job to learn all the idiosyncrasies of your office, ML algorithms tend to work better as they receive more data.
Why should human resources professionals pay attention to machine learning?
[Walsh] In the last decade, HR professionals have seen the amount of data they oversee proliferate. Data elements around skills, staffing, performance, payroll and many other areas have grown both in complexity and size. With all of this data it is getting harder and harder to simply view your data in a spreadsheet and quickly find the problems, let alone the solutions.
Machine learning holds the promise to unlock the hidden value in all this data that can no longer simply be perused by hand. For instance, machine learning can uncover patterns, such as those that might indicate compliance risk, and surface outliers that otherwise might have flown under the radar.
Similarly, the patterns detected by machine learning can uncover the root causes of key outcomes like turnover or daily business volume and even predict how these key performance indicators (KPIs) will trend in the future. All of these are use cases that HR professional have confronted for decades, but the growth of Big Data means machine learning tools are now capable and even needed to confront these familiar problems.
We keep hearing stories about how robots are going to replace humans. Can you give readers an example of how can machine learning can actually help people do their jobs?
[Walsh] No one is suggesting you turn your HR functions over to a robot, but there are plenty of areas where machine learning can help make manual tasks more efficient and provide insight that a human staring at rows and rows of data might have missed.
A great HR example is around compliance auditing. While many compliance regulations are enforced by rules in human capital management (HCM) and workforce management systems, more subtle violations are often lost in a haystack of Big Data.
For instance, small changes to an employee’s timecard might result in reduced overtime or meal break violations. Such changes may be nuanced, affecting only a few minutes at a time, and unfortunately there are hundreds of thousands or even millions of small edits in most organizations, most of which are not violations. Investigating them all by hand is not practical, but machine learning algorithms can uncover needles in this haystack and determine locations, departments, or employees that show abnormal patterns indicative of compliance violations. Not only does this make the auditor’s job easier, but it helps individual employees by ensuring fairness in time, attendance, and pay policies as well.
For organizations that are thinking about bringing machine learning into their workplace, what are 1-2 things they need to consider?
[Walsh] Machine learning is not one single algorithm or even one particular class of algorithms. Instead, there are many kinds of ML techniques, including:
- Unsupervised learning to uncover patterns
- Supervised approaches to predict outcomes for new data points
- Other variants like reinforcement learning
Ensuring success with these technologies starts by building the right business process around their use. Specifically, before bringing in any ML solution, you need to clearly define the business problem you are trying to solve and the metrics for measuring success.
Defining the business problem, such as ‘find potential compliance violations’ or ‘predict future business volume’ is the key to finding the right variant of ML to deploy. For instance, if you need to predict business volume in the future based on previous trends, you have a classical supervised ML regression problem. While data scientists and other experts can help you wade through the jargon of ML terminology, they won’t find the right solution unless you state the business problem concretely and set down clear success criteria.
Are there any potential downsides that organizations should discuss before bringing machine learning into the workplace?
[Walsh] For many workers and managers, the terms ‘artificial intelligence’ and ‘machine learning’ can evoke mixed feelings. While there is certainly promise that routine tasks could be automated by AI and ML, there can also be confusion about what a new system using these technologies will actually do.
A recent study from the Workforce Institute found that employees were excited to embrace the benefits of AIbut a lack of communication from employers about its impact left them feeling apprehensive. For many of these employees, it seems that clear communication from employers about what they expect an ML system to do and why it is necessary can go a long way to allaying these concerns. As with the introduction of new technology, communication and training make the process much smoother.
How do you see machine learning evolving over the next 3-5 years?
[Walsh] The rising use of ML is part of a growing trend towards allowing data to drive previously hard-wired processes. As an example, consider the problem of forecasting business volumes.
Decades ago, a person might stare at a spreadsheet and try to determine how many transactions a business would do a week from now. Later, static formulas were derived that made this process more automatic. Data from previous weeks and years were fed to a hand-crafted computer program and it predicted the number of transactions. But now a machine learning system like the one used in Workforce Dimensions can make forecasting even more data driven and reliable.
The ML system does not just feed data to a formula, but instead constructs the formula itself based on historical data, determining the factors and trends that are most predictive of your business.
The future for ML in the HR space is full of use cases that are ready to make this next leap in data driven innovation. While there will always be a human in the loop for these systems, over the next 3-5 years more and more areas of HR and workforce management will start to harness the power of their data to undergo similar transformations.
A HUGE thanks to Dr. Walsh for sharing his expertise with us. As you can see, we’ve only just scratched the surface in this conversation about the capabilities of machine learning.
As HR pros, we play a key role in designing work, which means we need to have an understanding of these technologies, so we can design work that’s meaningful for employees and valuable for the business.
Image captured by Sharlyn Lauby after speaking at the Flora Icelandic HR Management Conference in Reykjavik, Iceland17