Machine learning is cresting the fresh wave of 2017 HR trends. Gartner research predicts algorithms will positively alter the behavior of over one billion global workers by 2020, while over 3 million people can look forward to “roboboss” supervisors. Yvonne Bauer, Head of Predictive Analytics at SAP SuccessFactors, sees machine learning becoming more widespread this year as part of HR’s steady progression from art to data-driven science.
“More companies will look into machine learning, moving from individual projects to actual products built into HCM suites,” she said. “Conversational interfaces like chat bots and natural language processing will emerge this year, allowing companies to change how workers interact with the system and derive insights from those activities, including what people are working on and how engaged they are.”
Some machine learning applications will directly support strategic business goals like recruiting, reducing turnover or increasing sales productivity. Others hold new promise for further cost-savings through HR self-service automation. According to Bauer, HR departments have to become machine learning literate with at least a basic understanding of the possibilities and limitations so they can ask the right questions to get the most out of products and experts. Here are four questions HR professionals can ask as they explore machine learning for their organization.
Question 1: Is this process well-suited to machine learning?
Effective machine learning requires well-defined, measurable outcomes and a critical mass of meaningful data.
“Machine learning works particularly well where there’s a very specific outcome you’re trying to predict and you can systematically collect large amounts of predictive data influencing decisions related to that outcome,” said Steve Hunt, Vice President of Customer Value at SAP SuccessFactors. “Staffing is the perfect example, which is largely about making one decision. Defined jobs like hourly, high volume positions lend themselves well to machine learning. If we hire this person will they stay more than a year, sell a certain amount, and so forth?”
Question 2: What are our specific, measurable outcomes?
While online activities are generating a big-data avalanche, harnessing it for the good of the company is impossible without well-defined outcomes. What’s more, there are legions of unknown predictors of outcomes in measuring a wide range of issues such as employee turnover, characteristics of high-performers and healthcare costs.
“Start by clearly defining the problem; don’t start with the algorithms and the solution,” said Hunt. “For instance, many things impact turnover beyond past behavior, and they aren’t always associated with the company, like life events or an economic downturn.”
Machine learning tools will increasingly support personalized career paths and training options for employees, much like ecommerce sites serve up tempting new offers based someone’s buying history. In fact, Greta Roberts, CEO at Talent Analytics Corp. links machine learning to the shift HR has to make from trend forecasts to individual employee predictions, including retirement timeframes.
Question 3: Do we have the data that will allow our machines to learn?
Machine learning is a closed loop data iteration channel. Bad data leads to bad decisions. The machine gets smarter as it obtains more data over time. It’s really that simple and complex. HR needs to work early and often with senior management and line-of-business to systematically collect and synthesize the input data that allows the machines to actually learn.
“We’re seeing machine learning going outside HR into the line-of-business targeting higher-volume transaction areas, such as finding out what kind of field reps generate more sales or which drivers have more accidents or call center performance,” said Roberts. “The machine learns when you feed the outcomes back into it over time.”
Question 4: Will our algorithms break any laws?
Don’t assume that algorithms are inherently unbiased, objective or fair. Cathy O’Neil, author of the book, “Weapons of Math Destruction,” cautioned against the wholesale replacement of human HR teams with algorithms.
“A lot of what HR does is repetitive, and companies want to increase quality and decrease costs,” she said. “It sounds like a win-win, but hiring algorithms and processes are highly regulated, and people haven’t spent enough time thinking about the question of whether these algorithms that are now a large part of the process are following the law which prohibits discrimination based on race, mental health and other factors.”
O’Neil advised companies to audit their algorithms for legality. “Stop using your algorithm unless you have evidence from an outside audit that it’s legal because your company will be on the hook for violations.”
Of course there’s a host other issues to deal with as machine learning emerges, including data privacy and security. One of the first steps for HR this year is separating the grand claims about machine learning from what’s really possible and beneficial for workers and their companies.
Follow me @smgaler