IBM CEO Arvind Krishna recently reported that the company is using AI agents to take over work that had previously been done by several hundred HR employees. Specifically, IBM’s “AskHR” agent has automated 94% of simple tasks like vacation requests and pay statements. As a result, IBM laid off members of its HR team and used the savings to hire employees in other areas like software engineering, marketing and sales.
Complex HR Scenarios
We know that AI can be an effective tool in streamlining processes and improving efficiencies. Routine tasks like time off requests, requests for copies of pay statements or W2s are the type of things that likely could be done just as well by an AI agent as a human.
But what happens when a vacation request is also a request for time off that could be covered by the FMLA? Employees do not have to specifically ask for FMLA leave but do need to provide enough information so the employer is aware the leave may be covered by the FMLA. Once an employer is aware that an employee’s need for leave may be FMLA-qualifying, the employer is obligated to provide information about the employee’s rights and responsibilities. Will an AI agent be able to tell when a request for time off is a simple vacation request as opposed to something that could trigger the employer’s obligations under the FMLA?
Similarly, what if an employee’s request for a copy of her last paycheck also mentions a dispute with her manager that could constitute harassment? Employers have a legal obligation to prevent and address harassment in the workplace. Employers must therefore investigate all harassment complaints, even if they seem minor. If a concern about harassment is substantiated, the employer must take appropriate action to stop the behavior. Do we trust AI to be able to pick up on nuanced statements that could later be construed as a report of harassment?
Some companies have been routinely using AI and machine learning to analyze employee responses to employee engagement surveys. This can work well if the survey consists of questions that require the employee to select from a list of required answers. However, if the survey contains questions that permit open-ended responses from employees, the tool needs to be able to identify responses that require human follow-up. For example, if an employee’s response mentions behavior that could be construed as discrimination, harassment, or retaliation, a human needs to intervene and investigate.
Considerations
If you are considering AI as a way to streamline HR processes, consider following:
- Is the tool “smart” enough to pick up on workplace nuances where human intervention is critical?
- Be prepared to provide human oversight to make sure that employees are using the tools for the processes for which they are intended.
- For tools that require employee input, design the tools in a way that requires the selection of specific inputs as opposed to open-ended input, which could become a way to report concerns or other issues that need to be addressed by a human.
