The Current State of AI in Compensation

Compensation teams are often seen as the "hard-skilled" cousin of HR. Given the vast amount of data they manage, it's no surprise that AI is a hot topic among these teams. While the potential for AI is recognized, many are still in the early stages of adoption, carefully weighing the benefits against security concerns and its impact on decision-making. We explored this AI interest with Swati Bakshi, Director of Compensation Programs, Operations and Systems at Roblox, discussing practical applications, concerns, and future perspectives.

When we asked Swati what she sees as potential AI use cases for compensation teams, several key areas emerged—market benchmarking, data aggregation, and managing major program processes such as compensation cycles. We dive deeper into each below.

Practical Applications of AI in Compensation

1. Market Benchmarking

For most companies, their compensation teams rely on market surveys to analyze how they pay against the market. Companies such as Radford, Mercer, and Pequity all provide millions of anonymized, aggregate data points from various industries on salary, bonuses, and equity to compensation teams—but it’s on the compensation analysts to interpret the data.

From Bakshi’s experience in the compensation field, she has observed analysts spend considerable time manually matching jobs, by comparing external survey descriptions with internal descriptions. This would be a good use case to leverage AI, where it performs the initial screening of job matches. By automating the matching process, analysts could substantially reduce the time they allocate to this task. 

This time savings could translate to more strategic work being completed for the companies and increase the effectiveness of the money spent on surveys, as companies could make ready use of these surveys sooner.

Bakshi also pointed out that AI has the potential to enhance market analysis outcomes by providing a foundation for compensation benchmarking. AI could help integrate and analyze market and internal data, leading to more precise initial assessments. Companies can then overlay the budget and incumbent compensation data to make recommendations for compensation targets.

2. Internal Company Data Aggregation and Insight Generation

Now, external compensation surveys are useful, but as Bakshi points out, within a company there are also thousands of HR data points that are constantly updating. Data in an HR information system, applicant tracking systems, cap table management software, or benefits enrollment system all could provide rich details on the health of a company’s compensation programs. The difficulty in using this data for teams, though, is constantly downloading, cleaning, and interpreting that data.

Bakshi believes that there is an opportunity for AI platforms to help bridge the gap between technology and data insights, enabling organizations to consolidate data, generate insights using algorithms, and address specific compensation management needs. Potential use cases include identifying employees at risk of leaving based on compensation data, flagging employees whose compensation deviates from the company's pay philosophy, and providing real-time tracking of spending trends across teams. This can lead to improved compensation decision-making, reduced turnover, and optimized talent management.

AI in platforms like Pequity, where they already integrate to HR tech stacks, could streamline the aggregation of data from these various HR tech stack internal databases to generate actionable insights for specific teams and locations.

3. Compensation Cycle Automation

For many HR teams, the main event of their year culminates in compensation or merit cycles. This is the time that the entire company collectively reviews employee pay and adjusts as needed based on market pay changes, company or individual performance, and promotions. Setting up one of these cycles can often take months.

To help manage and scale their compensation cycles, companies will use software such as Pequity, shortening their build times and offering users a better UX to plan in. Bakshi notes her excitement over some of Pequity’s beta-AI offerings, which include creating formulas from real language.

As more companies incorporate AI into their compensation cycles, we as practitioners can help train the models by providing feedback and iterating on the process. Rather than having compensation analysts write or code the formulas for compensation cycles, there could be an opportunity to have the platform generate the code and set up compensation cycles, based on a set of defined compensation principles in natural language. As they say, you have to learn how to crawl before you start walking and running - similarly this will require significant investment (time, money and resources) before it starts bearing fruit.

At Pequity, we've long recognized the potential of AI to transform compensation processes by automating cycles, generating proration formulas, swiftly answering manager queries, and offering a comprehensive view of employee compensation history.To learn more, visit our website!

The above were just a few of the areas that Pequity and Bakshi dove into, but there were plenty of other potential future applications of AI in compensation, including:

• AI-powered chatbots that can answer manager queries about compensation policies and employee-specific situations. A chatbot in the form of a compensation analyst that can answer live employee specific questions as you ask them - like why did Sally get a higher increase than John last year?
• Automated generation of compensation cycle instructions and formulas.

Bakshi highlights the potential of utilizing AI to uncover efficiencies, particularly for small teams, by exploring how data collaborations could augment their abilities without the burden of self-sufficient data collection.

Challenges and Concerns

1. Bias in AI Models

AI is transforming compensation management, but it's not without challenges. Bakshi highlights that one major concern is AI's potential to inherit historical biases in compensation data. For instance, if certain groups have historically been underpaid due to gender or racial bias, AI models trained on this data may continue those trends, unintentionally perpetuating pay gaps. A study by MIT found that biased AI systems can lead to uneven pay recommendations, disproportionately affecting women and minorities.

In compensation, this could result in inconsistent salary bands, uneven performance-based pay, or inequitable promotion practices. While some argue that AI can be trained to reduce bias, it's critical to validate these systems thoroughly and ensure they are constantly monitored. Regular audits, transparent algorithms, and clear compliance frameworks are necessary to prevent legal risks and ensure AI-driven compensation decisions are equitable.

AI excels at streamlining standard processes that are repetitive in nature, but it struggles with exceptions and complex scenarios. Consider situations like compensation adjustments due to an international uplevel transfer to a different job family, one-time salary adjustments due to an expanded role, or geographic pay differentials for remote employees. These require a deep understanding of both company policies and individual employee circumstances—areas where AI is often limited.

In Bakshi’s view, this is why AI isn't poised to fully replace compensation professionals. Instead, AI should complement human expertise, automating routine tasks and freeing HR professionals to focus on nuanced decisions, such as resolving pay disputes or designing compensation plans for high-performing employees. By integrating AI into compensation functions, organizations can gain efficiencies without sacrificing the human element crucial for fair and balanced decision-making.

Conclusion

As AI continues to evolve, its role in compensation practices is likely to grow. While it presents exciting opportunities for efficiency and data-driven decision-making, it's crucial to approach AI implementation thoughtfully, considering potential biases, legal implications, and the need for human oversight in complex situations. Swati Bakshi emphasizes that AI's application in compensation is still in its early stages but acknowledges its potential to support compensation practitioners. The notion of AI fully replacing human involvement in compensation, where bots solely determine salaries and eliminate the need for human resources, is unrealistic.

This underscores the importance of balancing AI capabilities with human expertise. By striking the right balance, organizations can create more efficient, fair, and effective compensation processes for the future. The key to successful integration of AI in compensation practices lies in viewing it as a tool to augment human decision-making rather than replace it entirely. As we move forward, the most effective compensation strategies will likely combine the analytical power of AI with the contextual understanding and judgment of experienced professionals.