Author: Dan Kuang 

As algorithmic technologies—such as Artificial Intelligence (AI)—rapidly advance, these innovations are increasingly finding ways to exert its influence on Pay Equity (PE).  As amazing as this technology can be, early adopters are signaling mixed reviews.[1]  Whether these tools prove to be helpful or harmful depend on many practical factors.  This discussion hopes to share some important considerations to help employers land on the winning side of the PE algorithmic equation.

The growth of automated PE analytics tools over the past decade has made the emergence of algorithmic PE models a natural progression.  The promise of quick and easy PE solutions has piqued many employers’ interest and as they seek consultative assistance to field and select vendors, several commonly-shared challenges emerge among employers:

  • They assume that “pay equity” is a single uniform concept.
  • They lack a firm understanding of what the algorithmic PE tool is actually doing.
  • They are unaware of the potential compliance requirements and legal implications of applying algorithmic PE tools to affect pay decisions.

This gap in understanding can lead to unnecessary expenditures at best, and increased legal risk at worst.  This paper aims to close that gap by providing employers with actionable insights and practical considerations.

Challenge #1:  Pay Equity is Not Monolithic

In concept, PE may seem like a fairly simple and straightforward uniform concept.  In practice, however, PE can take on different forms.  While many people think of PE from a legal[2] and social justice perspective—“equal pay for equal work”—it can take on other forms.  For example, a compensation expert may conduct PE analyses to ensure that pay is not excessively above or below market reference–that compensation is competitive relative to the open job market.

For each of the many different flavors of PE, there are different algorithmic PE solutions available to help.  Consequently, an employer who wants to automate their Title VII -oriented PE analyses should avoid labor market benchmarking PE tools.

While this may seem confusing, the good news is that PE studies largely come in one of two general flavors:  1) Internal PE and 2) External PE -analysis.  Each will be discussed in detail in the next section.

Internal PE Analysis

Internal PE analyses are focused on employees WITHIN a company.  The focus of the PE study is to ensure pay equality among similarly situated employees.  The “equal pay for equal work” requirement is highly regulated and litigated (see e.g., footnote 1).  More specifically, employers are required by law to ensure pay equality among similarly situated employees.[3]

Classically, internal PE studies involve regression-based analytics measuring for pay differences along protected classifications (e.g., gender, race) after controlling for legitimate explanatory factors (where available).

External PE Analysis

External PE analyses are focused on pay equality/comparability BETWEEN companies.  These studies are conducted to ensure that a company pays its employees comparably relative to the external labor market.  In general, the decision to pay above, below, or level with the labor market is driven by corporate strategy.  It is argued that under-paying employees relative to the market would motivate them to seek higher paying jobs on the open market and create a turnover problem.

Unlike internal PE, employers are not required by law to ensure that their employees are paid at level with the labor market—external PE.  The understanding is that external PE will naturally be regulated by open-market pressures, i.e., employees who are underpaid relative to the market will leave to seek jobs that pay more.

Challenge #1 Summary

The big takeaway from this section is that internal and external PE studies differ in the problems they solve.  Because internal and external PE studies address fundamentally different needs and objectives, their respective analytical methods are notably different.

Challenge #2:  Algorithmic Pay Equity Tools

Despite the vast difference in analytic methods between internal vs external PE studies, their methodologies share a common philosophical roots—optimize analytic models towards a desired goal.  This section will explore:  (1) the specific objectives the algorithms are intended to achieve (“what”), and (2) the methods by which the algorithms aim to accomplish these objectives (“how”).

Algorithmic PE Tool–Internal PE

For internal PE studies, the goal may be to maximize the explanatory power of the analytic model to minimize between group pay differences (pay gap).  The range of algorithmic solutions to accomplish this is broad.  To put this into perspective, optimization algorithms relevant to internal PE studies date back to 1960 with stepwise regression.[4]  Between then and now, science has accelerated and analysts need help deciding which option is best for them.

On the “simpler” end of the spectrum, regression-based models are solid and robust solutions.  Optimized regression-based models are highly desirable because they are “white-boxes”:  each explanatory factor is clearly identified along with their contribution to the explanatory model.  Moreover, regression-based models have been widely relied on in PE litigation and are well recognized by the courts as reliable.

Contemporary non-linear AI methods offer the potential to increase explanatory power of analytic models and further narrow pay differences.  Although these qualities present well on paper, there are potential challenges that analysts may want to consider.  To begin, AI-based models tend to be “black-boxes” that are not easily understood nor accessible; it is not something that can be easily explained to the average person.  In application, AI-based models have yet to be pressure tested in the courtroom.

Experienced analysts understand that there is no one universally appropriate tool for all internal PE problems; their experience allows them to select the right (combination of) tools for a given problem.  Additionally, they understand the importance of grounding their analytics to a rational story that makes sense over time.  For that reason, experienced analysts have historically practiced what AI experts currently advocate—”keep humans in the loop”.

Algorithmic PE Tool–External PE

For external PE studies, the goal may be to identify the optimal salary that balances labor market compensation rates against the tension between workforce supply and demand.  The range of algorithmic solutions to the external PE problem is incredibly broad.  In its basic form, identifying optimal salary under specific constraints is a type of mathematical problem solved in 1939 by Kantorovich—linear programming.[5]  Advances in computational science, machine-learning, and AI have taken that science into the world of “dynamic pricing” where algorithms optimize price to market condition in real time.  This is the underlying science of Uber’s “Surge Pricing” and how airlines adjust seat prices as a function of supply and demand and other constraints (e.g., fuel).

While it may be tempting to solve all external PE problems with advanced dynamic pricing approaches, there are very good reasons for analysts to apply more modest approaches.  To start, simple linear programming methods are “white-boxes” where all model parameters are generally obvious and understandable.  In practice, compensation decisions need to be reliable and reasonably stable.  True dynamic pricing approaches can be overly sensitive to market conditions.  For these reasons, the experienced analysts approach external PE studies with an eye for the right balance of tools that will offer the most robust and rational solution for storytelling.

Challenge #2 Summary

Analysts have a very broad range of algorithmic solutions for their internal and external PE studies.  There are natural tensions between understandability and accessibility and algorithmic complexity.  A key takeaway from this section is for analysts to consider multiple options with a focus on delivering the most robust, reliable, and rational solution.

 

Challenge #3:  Compliance Requirements and Legal Implications

Decisions affecting compensation are highly consequential.  Naturally, there are compliance and legal guardrails to protect employees.  This section will explore practical considerations of algorithmic PE from a compliance and legal perspective.

Practical Considerations–Internal PE

The market for algorithmic internal PE tools is broad and extensive.  In general, these products are promoted as easy to use, therefore providing employers with convenient on-demand access to pay equity assessments.  Arguably, greater PE monitoring and awareness should help ensure greater PE, right?  Not exactly.  Outcome depends on many factors (e.g., data quality, comp practice, workforce dynamics), but to keep things simple, our recommendation lean on two primary factors: (1) analyst Experience, and (2) PE tool Automation.  Here is a recommendation heat-map where RED=Avoid and GREEN=OK.

There are two key takeaways from this heat-map.  First, experience is critical to outcome quality.  PE tools are so user friendly that an analyst who has never worked on a PE can run hundreds of thousands of analyses and obtain colorful dashboard outputs and pay adjustment schedules.  This is not recommended.  The analyst must have sufficient experience to properly understand the statistical model so they can evaluate its accuracy.  It is not uncommon for my analytic recommendations to disagree with PE tool outputs.  As an experienced analyst, I review and adjust the model to ensure it makes rational sense.  Algorithmic PE tools lack the necessary “experience” (aka, context) or the ability to evaluate if the model “makes sense”.

Second, highly automated PE may not inherently guarantee reliable and accurate outcomes; “over-automation” can, ironically, introduce unexpected challenges.  On the one hand, it can be helpful because highly automated PE tools can minimize manual effort and specification requirements.  On the other hand, however, highly automated PE tools take the human out of the loop, especially where experienced human judgement is most needed.  For employers shopping for an algorithmic PE solution, this is the biggest challenge as they seek to find their Goldilocks zone.  It bears emphasizing that the only assurance for reliable outcomes, regardless of PE analytic tool quality, is an analyst who is reasonably experienced with compensation practices and pay equity analytics.

Legal Exposure Considerations—When PE analytics are conducted outside the context of attorney oversight, the results are most likely not covered under attorney-client privilege (ACP).  This is not a trivial matter because ACP can shield PE analytics in related litigation disputes.  Automated algorithmic PE tools can further amplify the potential legal exposure if a company conducts many rounds of PE analyses—everyone of those PE studies can be subpoenaed for production.  This is particularly concerning because:  (1) PE problems that persist over time may be used to support a “pattern-and-practice” argument; and (2) highly dynamic workforce changes can artificially create ups and downs in PE outcomes, and inaccurately portray a compensation practice in disarray.  For these and many other reasons, it is strongly recommended that employers obtain sound legal advice before they implement automated algorithmic PE tools.

Pragmatic Considerations—In concept, “on demand PE analytics” is intuitively appealing.  In practice, however, there are important questions that employers need to address so they can safely implement this technology.

As a start, employers need to consider how often they will conduct PE study and support that answer with good reasoning.  PE is fluid and can go up/down due to dynamic personnel events (e.g. hiring, separation events).  It is important for employers to avoid unnecessary PE adjustments that would naturally normalize itself during the compensation merit process.

Equally important, employers need to consider how they will respond to identified PE problems.  As best practice, we work closely with clients to drill-down with root-cause investigations.  These follow-up investigations require considerable time and resources.  Will an employer have the resources to conduct proper follow-up analyses?  If they cannot commit the resources to properly respond, then what do they hope to gain?

While AI holds the promise to revolutionize PE, its “black-box” approach requires employers to place a lot of faith and trust that the algorithm, methodology, and execution are free of potential bias.  What efforts have AI-based algorithmic PE tools taken to assure their tools have been thoroughly evaluated for bias?  It is surprising how often clients implicitly trust that AI-based PE tools are free of bias.  As with all AI-based technology, the general wisdom is, “trust but verify”.

Practical Considerations–External PE

External PE studies are completed by trained compensation experts.  Algorithmic external PE tools automate and simplify tedious tasks and calculations for the experienced compensation analysts.  Having an experienced analyst in the loop significantly reduces the potential concerns for external PE tools.

As AI embeds itself into external PE tools, however, the technology introduces potential concerns that employers need to be aware of even with an experienced compensation analyst in the loop.  First, AI-models can evaluate and integrate data and provide suggestions for decisions typically requiring experienced compensation analysts’ judgment.  Effectively, the AI-model can influence the compensation analyst’s decision making.  At scale, this influence can exert measurable effects if there is bias in the algorithm.  What assurance is there that the AI algorithm is free of bias?

Second, AI-models are so powerful at integrating data that they can accomplish analytics that are unimaginable with humans in the loop.  For example, automated computational models can factor in real-time labor market supply and demand data, which brings “dynamic pricing” technology into the compensation setting space.  When paired with the fact that external PE tools rely on one commonly shared library of compensation salary data across multiple clients, a potentially concerning scenario emerges.  The dynamic pricing model is providing similar compensation recommendations across different employers, which has the potential to unintentionally coordinate wage and reduce salary competition among different employers.  For some, this may be considered as a form of collusion.

Challenge #3 Summary

Algorithmic PE tools can carry considerable compliance and legal exposure concerns given that they may influence highly consequential decisions affecting compensation.  Keeping humans in the loop can mitigate those potential exposures but it is no guarantee especially when algorithms provide suggestions that involve too many data points for the human to process and independently verify.

Overall Conclusions

While these challenges may all seem concerning, it is important to remind ourselves that they are more opportunities than they are threats.  Algorithmic PE tools are incredible options that are available to employers, who are not required to implement unless they feel they are ready.  To assist employers in their self-evaluation, this paper discussed three general questions that employers need to be comfortable answering:

  • What type of PE problem am I investigating and solving for?
  • What are the underlying analytics that the PE tool needs to perform?
  • What are the potential compliance and legal exposure that the PE tool may introduce?

Given the complexity of PE and algorithmic analytic tools available, it is impossible to catalogue all concerns and questions.  However, the hope is that this discussion helps employers to think about some of the more fundamental and basic questions so they can be informed consumers of this amazing technology.

[1] e.g., fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/ referencing nanda.media.mit.edu/ai_report_2025.pdf

[2] e.g., Title VII of Civil Rights Act of 1964 (42 U.S.C. § 2000e et seq.); Equal Pay Act of 1963 (29 U.S.C. § 206(d))

[3] Internal PE can be further sub-divided into two different methods of grouping employees for pay comparison:  (1) substantially similarly situated and (2) similarly situated.  These subtle and nuanced differences can significantly affect employee pay comparison composition.  This is outside of scope for the present paper, but the author is happy to discuss offline:  dkuang@resecon.com.

[4] Efroymson, M. A. (1960). Multiple regression analysis. In A. Ralston & H. S. Wilf (Eds.), Mathematical methods for digital computers (pp. 191–203). John Wiley.

[5] Kantorovich, L. V. (1939). Matematicheskie metody organizatsii i planirovaniya proizvodstva [Mathematical methods in the organization and planning of production]. Leningrad State University Publishing House.