‘AI Should Be Human-Led’: Dr Rushikesh Borse On Workforce Analytics

Traditionally, employee turnover prediction was done with conventional Machine learning (ML) algorithm, where data of previous 3-5 years for all the employees was learned by ML algorithms and based on various parameters the prediction percentage was computed. Off late, Data Scientists started using job satisfaction review text as an added feature to understand the subjective feelings of employees to comprehend the feeling of mind of the employee. This helped in better prediction. In recent times, along with the existing two methods, employers are leveraging AI enabled Video analytics for understanding employee work efficiency, workplace satisfaction, relationship with peer colleagues and overall mood of the employee. This technique estimates many aspects of the employee without compromising their trust and privacy.

AI is now part of almost every workplace decision, and that is not going to change. But there needs to be a clear line between what AI can do on its own and what needs a human to approve. 

A simple rule works well: AI can flag, suggest, and rank — but any decision that affects a person's job, pay, promotion, or role should always need human sign-off. For example, AI can highlight that an employee's performance has dropped, but a manager should be the one deciding what happens next. Lower-stakes work, like generating reports or spotting trends, can run with less human involvement. The line should be drawn based on impact: the bigger the impact on a person's career, the more human judgement should be involved. 

AI systems learn from past data, and if that data has bias; for example, favouring a certain age group or background, the AI can carry that bias forward without anyone realising it. So organisations need to regularly check their AI scoring systems: do the outcomes differ unfairly across gender, age, or tenure? These checks should happen often, not just once. 

Explainability also matters. For high-stakes decisions like promotions or performance ratings, companies should use AI models where the reasoning can be understood and questioned, not "black box" systems no one can explain. And there should always be a clear human owner responsible for the outcome, not "the algorithm decided." If an employee questions a score, someone accountable should be able to explain it and correct it if wrong. In short, the HR systems can be AI backed but human led, as the name ‘Human’ is in their role (Human resources).  

You see, predictive analytics and AI is leveraged by the HR team right from resume shortlisting, key skillset shortlisting, framing personal skill based interview questions. So, why not further in the process of decision making as well. Based on project description and client expectation, the HR team can take an AI opinion - ‘Ideal Team for this project’. Certainly things change as per company’s resource availability which needs to be human verified. 

Instead of thinking in terms of whole jobs disappearing, organisations should break roles down into tasks. Some tasks are repetitive and data-heavy, these are good candidates for AI to handle. Other tasks need human judgement, creativity, or relationship-building, these stay firmly human. Roles should be redesigned around this split, rather than assuming an entire job will vanish.  

For this, the First thing is that the organization should give an ‘AI literacy course’ to everyone. It can be a mandatory 10-20 hours certification course. An exam at the end of the course would be necessary to check that the course was taken seriously. Second, it needs to be told to the employees that AI is not going to eat their job, rather it would make their job easy, so better embrace and learn it for better productivity. Since, it is a competitive world, not comfortable with  AI would be a - fear of missing out (FOMO).