While AI definitely has its perks, there are certainly risks when bringing it into compensation work. Many compensation teams are cautiously curious about how AI could save them time, but also nervous about exactly how to bring this new technology into their existing processes.

A significant risk of AI adoption is the undervaluation of human expertise, which could lead to poor compensation decisions. AI, especially large language models (LLMs), can generate sophisticated-sounding but incorrect information. LLMs often predict how to finish sentences based on consumed words, rather than engaging in true legal reasoning or understanding principles. That means that if inaccurate information gets plugged in, inaccurate information comes out… and we all know that working with inaccurate information can lead to massive time-wasting and more work than it’s worth. 

For compensation teams looking to leverage AI, we recommend these 3 steps for getting started:

1. Understand Deterministic vs. Probabilistic Calculations.
Before introducing AI, clearly define which analytical functions should always remain deterministic (produces fixed, expected outcomes) and which functions can rely on more fluid outcomes. Identifying which type of calculations are necessary will help you determine which models are going to be most impactful for your team.

2. Establish Clear Governance. 
Collaborate with your IT and security teams to implement simple governance practices and principles, specifying when and how large language models should or should not be used. Next, provide education for all users on basic IP law and the risks of inputting sensitive or proprietary data into public AI models, as this can lead to serious legal consequences and security risks.

3. Recognize Current Usage and Build Data Literacy. 
Evaluate the data literacy and expertise of individuals who will be using AI systems, especially for high-risk applications that affect compensation and livelihoods. Acknowledge that employees are likely already using AI tools (e.g., ChatGPT) despite policies that attempt to deter their usage. Instead of an "on-off switch" approach, build up your team’s basic data literacy and understanding of how these tools work.

Like any new strategy, tool, or system, intentional implementation is key to actually getting the outcome you want. For AI use, maintaining data safety and correct usage is key to getting all the perks of the technology while still mitigating the risks.