Artificial intelligence (AI) is increasingly being used to analyse salaries, benchmark roles and assess pay equity, but employers should be wary of allowing technology to make remuneration decisions on their behalf.
This is according to Dr Mark Bussin, executive committee member at the South African Reward Association (SARA), who points out that AI is already changing the way organisations approach remuneration. From analysing large volumes of salary data to identifying anomalies and testing for pay-equity patterns, AI can help reward professionals work faster and spot trends that might otherwise be missed.
But Dr Bussin says there is an important distinction between using AI to support remuneration decisions and allowing it to make those decisions.
“AI can bring enormous value to reward teams, but when the outcome affects what an individual earns, organisations cannot simply accept what the system produces. There still needs to be human judgement behind the decision.”
An AI-generated number is not a salary benchmark
One of the biggest concerns is the quality of the information AI relies on. Publicly available salary data can be inconsistent, outdated or based on different definitions of remuneration. Job titles can also be misleading, with the same title representing very different levels of responsibility across organisations.
“An AI tool can give you a salary figure very quickly, but speed does not make it a reliable market benchmark. Employers still need credible remuneration data and properly matched jobs to make informed pay decisions,” he says.
The same applies to job grading. While AI can analyse information in a job profile, it cannot necessarily determine whether that profile is accurate or fully understand the complexity, responsibility and organisational context of the role.
The risk of repeating historical inequalities
There is also a risk that AI could reproduce existing inequalities if it is trained on historical remuneration decisions.
“Historical pay decisions reflect the practices of the past, including inequalities that may already exist. If those patterns are fed into an AI system, there is a risk that the technology will reproduce them rather than identify them,” says Dr Bussin.
This is particularly relevant in South Africa, where employers continue to focus on pay equity and fair remuneration. Simply removing factors such as race or gender from an algorithm does not necessarily eliminate bias if other variables act as proxies.
Keep the human judgement in the equation
Dr Bussin says employers should distinguish between three separate questions when making remuneration decisions: What is the external market paying? What is internally equitable? And what is appropriate for the individual, taking performance, experience and scarce skills into account?
“AI can help inform all three questions, but it should not turn them into one unexplained answer. Someone still needs to understand the methodology, challenge the result and take responsibility for the final decision,” he says.
Reward the value created by AI
AI will also change the nature of work itself, potentially reducing some tasks while increasing the importance of new skills and capabilities. This could require organisations to rethink how they structure jobs and reward employees.
However, Dr Bussin cautions against simply paying a premium to employees because they have access to AI tools.
“The real reward should be for the value people create with AI, through better judgement, higher productivity, innovation, risk reduction or improved service,” he says.
Ultimately, AI can make reward processes faster and more analytical, but it cannot remove the need for sound remuneration practices.
“Trustworthy reward decisions still require credible data, a sound job architecture, transparent governance and experienced human judgement,” says Dr Bussin.