Flawed Expectations: Unpacking the Potential of AI in HR

AI has stormed the gates of HR, promising a future of hyper-efficiency and minimized bias across the entire employee lifecycle from recruitment to retention. Yet, a shadow looms over this technological promise. The International Labour Organization (ILO) warns that the rapid deployment of AI in HR is barreling into substantial empirical and ethical limitations.

The industry consensus is clear: the challenge is the flawed expectation that this technology can revolutionize HR without HR first revolutionizing itself.

The Limits of Automation: Risking Inequity

In its report, “AI in human resource management: The limits of empiricism,” the ILO delivers a stark warning: AI risks perpetuating inequities rather than resolving them. While HR tech providers rightly celebrate their efficiency gains, the industry is navigating the challenge of also addressing deeper, more complex ethical considerations.

Data from OCC paints a picture of aggressive adoption in key markets like Mexico. Following planned deployments, an estimated 36% of companies are now actively integrating AI into their core HR processes. AI applications are heavily concentrated in areas that demand high-volume data processing, such as talent acquisition and administrative automation.

Top AI use cases in Mexican HR include automated candidate screening (44%), CV and interview analysis (38%), and automation of repetitive administrative tasks (32%).

The accelerated adoption of AI is driven by its unmatched ability to process massive datasets in real time, identify decision-support patterns, and slash operational workloads. As HR teams grapple with increasing complexity, AI is shifting its identity from a simple efficiency tool to a strategic enabler, forcing a crucial choice for leaders: using AI to simply do old HR tasks faster, or fundamentally redefine how to value and connect with human talent.

ILO's Warning: The Limits of Empiricism

The ILO framework identifies three core structural limitations that challenge the credibility and fairness of AI systems:

AI often attempts to quantify human qualities that are inherently complex. This tendency toward reductionism is a core ILO concern. The report highlights cases in candidate selection where algorithms applied “reductive or exclusive” criteria, such as rejecting a qualified engineer because their degree was in Arts rather than Science, despite their credentials.

"AI can reduce the time tremendously, but whether it does so by achieving the objective of selecting the best candidate is less evident," writes the ILO.

This restrictive focus might limit organizations from finding the candidate that best suits their needs. “AI enhances human capabilities, but it cannot capture the full spectrum of creativity or leadership,” says Leopoldo Ocaña, CEO, Fleet. 

Low-Quality or Biased Data: Automating Discrimination

Data quality is a second critical limitation. AI systems inherently inherit the patterns present in their data, meaning historical discrimination is encoded into the prediction. The ILO cites multiple examples of algorithmic bias:

• Recruitment and Sourcing Bias: A Facebook ad experiment revealed that job advertisements for STEM roles were more frequently shown to men, because the algorithm learned that targeting men was cheaper, even though the ad itself was gender-neutral.

• Historical Discrimination: Algorithmic targeting perpetuated stereotypes, showing women ads for “secretary” or “nurse” positions while men received STEM or construction job postings.

• Compensation Bias: Gig-economy systems sometimes identify groups willing to accept lower wages and then repeatedly route lower-paying tasks to them, suppressing income potential in markets where frontline workers earn around MX$8,400 (US$459) per month.

In Mexico, representation gaps remain sizable. Informal workers (over 50% of the workforce), women in frontline roles, and candidates in rural or peripheral regions are consistently underrepresented in digital datasets. These structural representation gaps create blind spots that algorithms can reproduce.

“AI bias is a risk; human bias is a certainty... In Mexico, where discrimination by age, gender, and appearance is widespread, algorithmic oversight combined with data audits can generate more diverse shortlists than manual screening,” says Vera Makarov, Co-Founder and CEO, Apli.

Some Mexican companies are expanding their approach to circumnavigate these limitations. For example, Mexican job platform Chambas AI says that it attempts to reduce exclusionary bias by using verified histories via WhatsApp, allowing candidates without formal documentation to demonstrate experience. It also prioritizes skill- and proximity-based matching over formal credentials, while proactively including operational workers whose fragmented histories typically push them out of algorithmic visibility.

This data suggests that efficiency gains do not automatically ensure equity; they depend entirely on data governance, not just algorithms.

Non-Transparent Programming: The Black Box Problem

A third structural limitation is the opacity of AI systems, often described as the “black box” problem. Even when algorithms improve efficiency, their underlying logic is rarely visible to HR managers, workers, or even the vendors who deploy them. According to the ILO, opaque programming can conceal embedded biases, prevent meaningful scrutiny, and reduce organizational accountability.

This challenge is visible in the Mexican ecosystem, where advanced predictive models are rapidly being introduced. For example, solutions like Fleet build turnover and performance prediction models, but users often see only risk flags, not the underlying variables or weightings. The challenge only deepens as systems become more autonomous. 

While automation is promising, it raises a central governance question: Who is accountable when the machine initiates decisions but the logic is inaccessible? Without standardized frameworks for Explainable AI, the sector risks undetected discriminatory outcomes, reinforcement of historical inequities, and legal exposure, as global regulations evolve.

This limits the applicability of AI in fields where transparency is a concern. Maya Dadoo, CEO, Worky, emphasizes the complexity: “Payroll in Latin America involves union rules, industry-specific tax schemes, and unique compensation models.” Worky’s zero-fine guarantee conveys compliance, but these systems rarely expose their algorithmic decision-making process regarding tax calculations or staffing optimization.

AI's Purpose vs. HR's Mandate

The analysis reveals that HR tech in Mexico is delivering tangible operational improvements, reducing turnover, saving administrative time, and accelerating job placement. Yet, the core risks flagged by the ILO persist: reductive objectives, biased data, and opaque programming.

The final message circles back to the starting tension: AI is transforming HR, but its true benefit will be determined by its governance. AI's inherent design goal is efficiency and complex pattern recognition; it is fundamentally a tool of optimization. Therefore, it must be acknowledged that achieving social equity is not an automatic function of the algorithm, but an external ethical mandate imposed by society and necessary for HR's legitimacy, flagged by the ILO. 

The evidence is clear: adopting AI without first improving data quality, representation, and oversight risks automating the very inequalities HR leaders are trying to fix, purely in pursuit of efficiency. The promise of AI will only be fully realized when it is intentionally paired with a commitment to fairness and transparency. 

AI should be positioned as a tool that “improves [job seekers’] chances of securing better jobs and working conditions,” says Max Werner, Founder, Chambas AI. AI can enhance productivity, but it must not reduce human potential to a set of data points.

Only by questioning assumptions, scrutinizing data, and rejecting the flawed expectation that AI alone can fix HR, can the technology fulfill its initial promise: transforming HR into a strategic enabler that empowers employees and organizations alike, without inadvertently entrenching the fundamental inequity it was meant to help resolve.