Sara Hillenmeyer, VP Of AI & Data Science At Payscale On Building Fair, Explainable, And Decision-Ready Compensation Intelligence

Sara Hillenmeyer: Most leaders do not struggle with a lack of data. They struggle with knowing which data actually matters. The first step is choosing a source they can trust.  Payscale’s 2025 Compensation Best Practices Report showed 52% of organizations use free or open online data to price jobs. However, HR also says this source is the least trusted. Data only becomes decision-ready when it has been cleaned, validated, and normalized, so leaders are comparing information that is reliable. Free, online data can be biased and shouldn’t be used as a single source of truth. For example, searching a salary range through Google or ChatGPT could be skewed. This is because tools like ChatGPT often pull data from broad, uncurated sources like Reddit, which might reflect states with pay transparency laws or larger organizations, and thus might not be representative of your location, industry, or organization size.

The 2025 CBPR also shows that organizations trust traditional salary data the most, followed by HR-reported data, like Payscale Peer. Why is this data more trusted? Because the data is verified and from validated sources.  Trusted data gives leaders a clear picture of what is truly happening in the market. Traditionally, the gold standard for compensation data has been HR-reported data sources, such as comprehensive salary surveys. These sources are highly valued because the data is collected directly from employers, offering a cleaned and normalized view of what similar organizations are paying. However, the labor market is moving quickly, especially for hot jobs where wages can rise by eight to ten percent in a short period of time. Leaders need data that reflects what is happening today, not data that is six or twelve months old. This is why next-generation HR-reported data solutions are gaining traction, providing the same high quality of traditional survey data but with near-real-time currency.

Once organizations identify a reliable data source,  the next step is focusing on the story it tells. Leaders should look for patterns that connect directly to business outcomes. They might examine how pay influences hiring speed or turnover. Even simple insights, such as whether the company is above or below a local midpoint, can guide better decisions. The value comes from reducing complex datasets into a small set of clear insights that support the company’s pay philosophy and talent strategy.

The final step is building a consistent process. Leaders need a framework they can use every year. Consistency builds trust in the data and in the decisions that follow. Without a structured approach, anecdotal information can take over, and compensation decisions can drift away from the organization’s strategy. A consistent approach builds trust with both employees and leaders. When they trust the process, confidence follows.

Sara Hillenmeyer: AI is very good at spotting early signals in the labor market. When a skill begins appearing more often in job postings or resumes, these trends often show up well before traditional compensation surveys capture them. This early visibility helps organizations anticipate talent needs and plan pay decisions proactively.

The challenge is interpreting these signals responsibly. Not every spike in demand represents a long-term trend. That is where human judgment is essential. Compensation and HR teams need to evaluate whether a new skill delivers meaningful value in their specific business context. AI provides the early insights, but fairness requires careful review and thoughtful weighing of the information.

By combining AI-driven visibility with human oversight, companies can stay competitive while avoiding overreactions or inequitable pay adjustments. This ensures emerging skills are recognized and rewarded in a way that supports both business outcomes and employee fairness.

Sara Hillenmeyer: Empathy begins with understanding the real impact pay decisions have on people. This is not an abstract exercise. Compensation affects financial stability, career growth, and personal well-being. With that in mind, we design AI systems with fairness checkpoints built in from the start rather than adding them later.

Bias reviews, model monitoring, and evaluations by domain experts are all part of the development process. We also continuously monitor model performance because compensation data shifts quickly. A model that worked well a year ago may not be accurate today, so ongoing review is essential for both fairness and accuracy.

Most importantly, human oversight never goes away. Algorithms can surface patterns and highlight opportunities, but they should not make final judgments about people. Technology is there to support more equitable decisions and inform human judgment without replacing it.

Sara Hillenmeyer: Explainability is essential for compensation work. Leaders must be able to understand why a recommendation exists, especially in an environment where pay transparency is expanding.

Historically, compensation professionals have been tasked with using their training and experience to make reasonable inferences in situations where there is no relevant market data to lean on by triangulating between the data points that they do have access to. We design models that perform this task using the same approach that a compensation professional would take, except with access to more data and computing power. The models show how specific factors influence compensation for each job. For example, the systems separate the effects of geography, industry, and company size instead of treating those forces as universal. This helps leaders see the reasoning behind a range rather than receiving a number without context.

Clarity helps people feel confident taking action. When leaders understand why a model produced a result, they can discuss it with employees, managers, or boards without relying on technical language. The goal is not only predictive accuracy but also communication that supports better pay conversations.

Sara Hillenmeyer: Responsible scaling begins with traceability. Any AI system that supports compensation decisions must allow people to understand how it arrived at its recommendation. This creates accountability. It also allows HR teams to evaluate whether assumptions still hold after market shifts.

We also monitor what is known as model drift. Compensation markets change over time, so models must be refreshed when they start to lose accuracy. This is not a one-time task. It is part of the long-term maintenance of any AI system.

Finally, responsible scaling means keeping humans in control. AI can increase consistency and help reduce bias, but pay decisions should always involve a person who understands the broader context of the business and the people within it.

Sara Hillenmeyer: Language models are making analytics more accessible. HR teams will be able to ask questions in plain language instead of relying on specialists who know SQL or programming languages. This will allow more people to investigate pay equity, review trends, and explore their data in ways that were difficult in the past.

The ethical considerations are significant. Employee data must be handled carefully, and models need boundaries that prevent overreach. Language models should summarize, categorize, and surface patterns, not make inferences about an individual’s potential or future performance.

When used appropriately, these tools can help HR leaders spend less time manipulating data and more time thinking about strategy and people. The key is strong governance and continued human judgment.