AI could help determine your salary, and that's a good thing

Artificial intelligence (AI) is expanding across all industries – including the compensation software space. This raises valid concerns among experts, especially given how new the technology is. How will AI make pay decisions/? Will AI place employees at a competitive disadvantage in negotiations? Will the use of AI cause greater pay disparity since it has to learn from previous data points that likely include biased decisions?

All of these are real anxieties. Over the last decade, many tech giants – Google, Facebook and Amazon to name a few – have attempted to use AI to run their pay programs; and each of these companies ultimately turned off their early AI models because they produced biased and undesirable outcomes.  I witnessed this firsthand as a team member at Google during one of their AI experiments. They tested the AI on candidate data to see if it could make compensation offers based on the data in Google’s HR systems – and they shut down the AI because outcomes were incredibly biased. The experiment shone a light on a problem that is still true for AI today: disparate pay outcomes resulted from the algorithm exponentially magnifying the bias that existed in all of the previous decisions leading up to a candidate getting to an offer stage at Google.

Bias in AI is real

See, bias exists in every structure that goes into society, meaning it’s infused in every step that leads to a person getting a job offer – the neighborhood someone grew up in has a socioeconomic bias; there is bias in the university’s selection process; there is bias in the way job descriptions are written, and therefore bias in which resume is picked up by the applicant tracking system; there is bias in the rating interviewers give a candidate; and finally, there is bias in how a manager selects the level they want to hire a person in at.

Although seemingly small instances on their own, once fed into a rules engine, each of these biases becomes part of a learning loop that the algorithm ingests and repeats over and over, until it’s less of a nuance, and more of a “rule” in the AI’s decision-making process.

This is a hurdle that compensation software using AI must overcome head-on, and it’s also why it’s so important for compensation companies, particularly those handling compensation data, to lean into the Artificial Intelligence trend.

Based on what I learned at Google, it may be easy to suggest that AI doesn’t belong in compensation data at all – still, even fear of change won’t stop it from happening. If there is any hope to curb AI from driving worse outcomes, we need fearless companies to take on the challenge and implement new ways for AI to help us pay better.

In truth, there are many manual pay processes that AI can take over from HR and compensation teams. Some examples include reading job descriptions and then more accurately matching roles to a survey; AI could be tasked with evaluating a company’s pay equity based on employee census data, and make recommendations for remediation; it could compare market data to employees and propose which ranges to adjust or maintain based on a company’s performance.

Essentially, if AI lives up to our expectations, it could perform many of the analyses HR and compensation leaders need in a fraction of the time. Most will agree this is an incredibly powerful benefit, however, knowing it could produce biased outcomes could slow industry adoption – especially if HR leaders plan to apply AI to determining people’s wages.

Enter predictive analytics: Enabling safer adoption of AI and more equitable pay

First, you must understand there is a distinction between predictive analytics and AI to truly understand how they can work together. Predictive analytics forecast future outcomes using inputs like statistics and historical trends; it can be part of an AI model, but unlike AI, it is not built to learn, reason, and self-correct. In other words, predictive analytics will always use the same assumptions to forecast an outcome, always asking the question, “What will happen next?” On the other hand, AI can correct its own assumptions and even explain this self-correction. In a Venn diagram, predictive analytics and AI overlap, though they are still distinct concepts.

Predictive analytics could enable safer adoption of AI in compensation for this reason: it can allow us to forecast the exponential impacts of what the AI is proposing. For example, you can ask AI to create new pay ranges based on updated market data. You can then forecast that range change across your organization over a one-, two-, or even five-year period. While it may be hard to see a single issue in an organization, over a longer time and a larger data set, those trends become more visible and easier to identify. For instance, if the predictive analytics flag that increasing a certain set of ranges over another leads to drastically underpaying a minority group in your company, you can take that assumption into account in your AI model.

By pairing AI and predictive analytics you can create both faster compensation outcomes and more equitable compensation outcomes. The pairing won’t be perfect to start, but eventually, you can even train the AI model to predict future outcomes on its own and self-correct its assumptions.

Related: Report: Online learners prefer human connection over AI

This all may sound like science fiction, and to some degree, it is. But the future isn’t that far off. Imagine a world where AI and predictive analytics work together to:

• More equitably level males and females, and predict leveling progression for the future, directly attacking the leveling discrepancy or “glass ceiling” in organizations.
• More accurately set a “minimum living wage” in states or regions where the federal minimum wage falls below the poverty wage, and track inflation to stay above the poverty line.
• Forecast equity pool size needs based on future granting and fundraising efforts, and identify inequitable outcomes based on current granting practices.

By combining AI and predictive analytics, you enable a more reliable, less biased algorithm that, over time, opens the world of compensation up to endless possibilities for a better future. It all starts with a massive amount of data, and a fine-tuned attention to not just how AI will benefit us now, but what it means for the future. What will happen next? I can’t wait to see.

Kaitlyn Knopp is the Founder and CEO of Pequity, the intelligent compensation platform that helps HR teams save their companies time, money and talent by catching compensation problems before they happen.