When a mid-sized professional services firm in Kitchener, Ontario programmed its automated pre-hire screening system to require 7 or more years of Canadian legal experience for 2 legal positions advertised in 2024, the algorithm executed its instructions with mechanical precision, filtering out every applicant who did not meet that threshold before any human being reviewed a single résumé. Among those filtered out was a 29-year-old lawyer called to the Ontario bar 18 months earlier, with credentials from a law faculty outside Canada. The system registered that this applicant had less than 3 years at the bar, determined the threshold was not satisfied, and generated an automated rejection. No one in the firm's HR department saw the application. No hiring manager reviewed the candidate's qualifications, assessed whether the international credentials demonstrated equivalent competence, or considered whether the 7-year requirement was genuinely necessary for the role. The algorithm did exactly what it was told to do, and in doing so, it may have accomplished something the firm never intended: it may have created a screening mechanism that systematically excludes candidates on the basis of characteristics protected under Ontario's human rights framework.
The phenomenon at the heart of this lesson is proxy discrimination, sometimes called indirect discrimination or adverse impact discrimination. It arises not from overt exclusionary intent but from facially neutral criteria that correlate with protected grounds and produce discriminatory outcomes. Understanding how experience thresholds function as proxies, why they create legal exposure even when designed without discriminatory purpose, and what distinguishes lawful job requirements from unlawful screening barriers is essential for any professional responsible for designing, deploying, or overseeing automated hiring systems. The Kitchener firm's situation illustrates how readily a straightforward-seeming parameter can transform a screening algorithm into a vehicle for systemic discrimination, and why compliance requires scrutinizing not just the face of a criterion but its practical operation.
Proxy discrimination operates through correlation. When a screening criterion happens to align closely with a protected characteristic, applying that criterion produces outcomes that mirror what would occur if the protected characteristic were itself the basis for exclusion. The criterion serves as a stand-in, a proxy, for the protected ground. In employment contexts, the most common proxies are requirements that appear job-related on their surface but that, when examined empirically, screen out protected groups at disproportionate rates without genuine necessity. The requirement may be framed in neutral language, such as a minimum number of years of domestic experience, but its effect may be to exclude candidates whose protected characteristics made it impossible or impractical for them to accumulate experience in the specified manner. The Human Rights Code does not require proof that the employer intended to discriminate. It requires only that the criterion has the effect of disadvantaging individuals or groups on the basis of a protected ground, unless the employer can demonstrate that the criterion is a bona fide occupational requirement.
Experience thresholds are particularly susceptible to functioning as proxies because the accumulation of Canadian professional experience is itself shaped by factors that correlate with protected grounds. Consider the 7-year threshold applied by the Kitchener firm. A person who immigrated to Canada 3 years ago cannot possibly have 7 years of Canadian legal experience regardless of their competence, diligence, or total years in practice elsewhere. A person who took an extended parental leave may have fewer Canadian years at the bar than a colleague of the same age who did not. A person who faced systemic barriers to licensure, perhaps because their foreign credentials required additional assessment or bridging, may have begun their Canadian legal career later than someone educated domestically. Each of these circumstances correlates with protected grounds: place of origin, ethnic origin, sex, family status. The 7-year requirement does not say "no internationally educated applicants," but in practice it may exclude them at dramatically higher rates than domestically educated applicants. It does not say "applicants who took parental leave are disfavoured," but it may penalize those who did. The neutrality of the language does not insulate the criterion from scrutiny under the Code.
The legal framework governing proxy discrimination under Ontario's Human Rights Code centers on the concept of adverse effect discrimination. This form of discrimination occurs when a rule, standard, requirement, or practice that is neutral on its face nonetheless has a disproportionate adverse impact on individuals identified by a protected ground. The Code protects individuals from discrimination in employment on the basis of race, ancestry, place of origin, colour, ethnic origin, citizenship, creed, sex, sexual orientation, gender identity, gender expression, age, record of offences, marital status, family status, and disability. When an employment practice screens out candidates at a higher rate based on any of these grounds, the employer must be prepared to justify the practice as a bona fide occupational requirement. The employer bears the burden of demonstrating, first, that it adopted the standard for a purpose rationally connected to the performance of the job; second, that it adopted the standard in an honest and good faith belief that it was necessary to the fulfilment of that purpose; and third, that the standard is reasonably necessary to accomplish that purpose, meaning the employer cannot accommodate individuals affected by the standard without incurring undue hardship. This tripartite test governs whether a criterion that produces adverse impact can nonetheless be maintained.
Automated screening systems amplify the risks associated with proxy discrimination because they apply criteria uniformly, at scale, without the individualized assessment that might identify exceptions or prompt accommodation inquiries. When a human reviewer considers a candidate, the reviewer may notice that an applicant has 5 years of Canadian experience plus 10 years of practice in another common law jurisdiction, and may pause to consider whether that background satisfies the genuine needs of the role even if it does not satisfy the stated threshold. An algorithm does not pause. It does not notice. It receives instructions and executes them. The 7-year threshold becomes a hard gate: applicants who meet it proceed; applicants who do not are rejected without further consideration. The Kitchener firm's system operated exactly this way. It did not flag the 29-year-old lawyer's application for human review despite the presence of foreign credentials that might bear on the assessment. It did not generate a report indicating that a potentially qualified candidate had been excluded. It simply moved the application to the rejection queue and generated a form letter. The automation did not create the criterion, but it enforced the criterion with a totality and rigidity that human review would not have achieved.
The risk is not merely that individual candidates are wrongly excluded. The risk is systemic. When an algorithm applies a proxy criterion across hundreds or thousands of applications, it produces a pattern of exclusions that has aggregate effects. If the 7-year requirement disproportionately excludes internationally educated lawyers, the firm's applicant pool will become progressively less diverse over time. The firm will hire from a narrower segment of the qualified population. The excluded candidates will face compounding disadvantage as they are repeatedly screened out by similar systems at other employers. The labour market becomes segmented along lines that track protected characteristics, and that segmentation becomes self-reinforcing. Systemic discrimination does not require a conspiracy; it requires only that neutral-seeming practices with disparate impact are adopted widely and applied without scrutiny. Automated systems are particularly effective at achieving this result because they can process volumes that would be impossible for human reviewers, thereby magnifying the impact of any embedded bias.
Understanding how a specific threshold functions as a proxy requires examining both the criterion itself and the population to which it is applied. The 7-year requirement imposed by the Kitchener firm would need to be assessed by asking several questions. First, what is the demographic composition of the applicant pool that satisfies the threshold compared to the pool that does not? If applicants who lack 7 years of Canadian experience are disproportionately recent immigrants, members of racialized communities, or members of other protected groups, the criterion produces adverse impact. Second, is there a genuine connection between having 7 or more years of Canadian experience and performing the job competently? The positions advertised were legal positions at a mid-sized professional services firm; it would need to be established that candidates cannot perform those roles competently without precisely 7 years of domestic experience, as opposed to some lower threshold or an alternative measure of competence. Third, could the genuine needs of the role be met by a different criterion that would produce less adverse impact? If a skills-based assessment, a competency interview, or a probationary period could identify qualified candidates without imposing a rigid experience threshold, the existence of those alternatives undermines the claim that the threshold is reasonably necessary.
The international credentials of the excluded applicant in the Kitchener scenario make the proxy function especially clear. A lawyer called to the Ontario bar 18 months earlier, with credentials from a law faculty outside Canada, may have extensive legal experience acquired abroad. The individual may have practiced for many years in another jurisdiction, developing expertise in areas directly relevant to the advertised positions. The 7-year threshold, however, measures only Canadian experience. It treats foreign experience as non-existent for purposes of the screen. This approach embeds an assumption that Canadian experience is qualitatively different from and superior to foreign experience, an assumption that may not withstand scrutiny if the underlying skills and knowledge are substantially similar. The assumption correlates directly with place of origin, ethnic origin, and potentially citizenship, because persons educated and trained abroad are more likely to hold foreign credentials. The Code does not permit employers to impose requirements that effectively exclude internationally educated professionals unless those requirements can be justified as bona fide occupational requirements. Credential recognition and prior learning assessment frameworks exist precisely because Ontario's policy is to facilitate the integration of internationally trained professionals, not to permit their wholesale exclusion through rigid domestic experience thresholds.
The automation of the rejection compounded the problem in this scenario. When a human reviewer rejects a candidate, there is at least the possibility of an inquiry. The candidate may request reconsideration, the reviewer may notice an anomaly and pause, or a supervisor may audit a sample of rejections and identify a pattern. When an algorithm rejects a candidate, none of these interventions occurs unless the system is specifically designed to trigger them. The Kitchener firm's system was not so designed. It operated as a closed gate, applying the 7-year requirement without exception and without generating any signal that a potentially discriminatory pattern was occurring. The 29-year-old applicant received a rejection notice, likely identical to the notices received by other rejected candidates, with no indication that a human being had ever considered the application. The absence of human review is not, by itself, a violation of the Code, but it removes the safeguards that might prevent or detect violations. It also raises evidentiary issues if a complaint is filed, because the employer will have difficulty demonstrating individualized assessment when none occurred.
Employers sometimes assume that because an algorithm applies a rule consistently, without regard to the protected characteristics of any particular applicant, the algorithm cannot discriminate. This assumption is mistaken. Consistency is not a defence to adverse effect discrimination. Indeed, the consistent application of a discriminatory criterion produces consistent discrimination. Every applicant who lacks 7 years of Canadian experience is rejected, and if that criterion disproportionately excludes protected groups, every rejection contributes to the adverse impact. The algorithm does not need to know the applicant's place of origin to discriminate on that basis; it needs only to apply a criterion that correlates with place of origin. The correlation does the work. The Code is concerned with effects, not intentions, and the effect of the Kitchener firm's threshold was to exclude a candidate whose exclusion correlated with protected grounds.
The question of whether a given threshold constitutes a bona fide occupational requirement cannot be answered in the abstract. It requires a fact-specific analysis that considers the nature of the job, the genuine needs of the employer, the availability of alternatives, and the extent of the adverse impact. For the 2 legal positions advertised by the Kitchener firm, the analysis would ask what tasks the successful candidates would perform, what skills and knowledge those tasks require, and whether 7 years of Canadian experience is genuinely necessary to possess those skills and knowledge or merely a convenient proxy for them. If the positions involve appearing before Ontario courts, knowledge of Ontario procedural rules is relevant, but that knowledge can be acquired in less than 7 years. If the positions involve advising clients on Ontario law, familiarity with Ontario statutes and regulations is relevant, but that familiarity does not require 7 years to develop. If the positions require judgment and professional maturity, those qualities are not unique to persons with 7 years of domestic experience. At each step, the employer would need to demonstrate that the threshold is not merely rational but reasonably necessary, and that no accommodation short of the threshold could meet the genuine need.
Automated systems present particular challenges because the criteria they apply are often inherited from legacy practices, adopted without analysis, or set by vendors who have no knowledge of the specific roles being filled. The Kitchener firm may have programmed the 7-year threshold into its system because a senior partner believed that was the minimum acceptable experience, or because a vendor's default settings included that parameter, or because no one specifically considered what threshold was appropriate. None of these explanations would constitute a defence to a discrimination complaint. The employer cannot shift responsibility to the vendor for criteria the employer chose to apply, and the employer cannot escape liability by pleading that no one thought about it. The obligation to avoid discrimination applies regardless of how the discriminatory criterion came to be embedded in the system. Employers deploying automated screening must therefore examine every criterion the system applies, assess whether each criterion produces adverse impact, and determine whether each criterion can be justified as reasonably necessary to the job.
The concept of a bona fide occupational requirement is not a loophole that permits employers to impose any requirement that relates in some way to the job. It is a narrow exception that permits requirements genuinely necessary for the safe and effective performance of the work. The requirement must be connected to the actual duties of the position, not to general assumptions about what kind of candidate is desirable. It must be adopted in good faith, not as a pretext for exclusion. And it must be the least discriminatory means of achieving the employer's legitimate purpose, meaning the employer must have considered and rejected alternatives that would produce less adverse impact. A 7-year experience threshold for a legal position would need to satisfy all 3 elements. The employer would need to explain what about the position requires 7 years rather than 5 or 3 or some other figure. It would need to demonstrate that it adopted the threshold based on an honest assessment of job requirements, not on convenience or tradition. And it would need to show that it considered whether the genuine needs of the role could be met by assessing candidates on competence directly rather than using years of experience as a proxy for competence.
The risks identified in this lesson are not hypothetical. Regulatory bodies, advocacy organizations, and academics have documented the discriminatory effects of experience thresholds in multiple sectors. Employers who rely heavily on such thresholds without scrutiny have faced complaints, investigations, and remedial orders. The automation of these thresholds does not reduce the risk; it increases the risk by ensuring that the threshold is applied without exception and without the moderating influence of human judgment. The Kitchener firm's situation exemplifies how a well-intentioned effort to streamline hiring can create systemic barriers that the employer never intended and may not even recognize until a complaint is filed. By the time the 29-year-old applicant challenges the rejection, the firm may have excluded dozens of similarly situated candidates, all through the mechanical operation of a criterion no one examined.
Recognizing the proxy function of experience thresholds is the first step toward compliance. The subsequent steps, which include designing override protocols and conducting adverse impact assessments, require understanding what makes a criterion function as a proxy in the first place. That understanding rests on the principle that discrimination law looks to effects, not just intentions. It rests on the principle that neutral criteria can produce discriminatory outcomes. And it rests on the recognition that automated systems, precisely because they are efficient and consistent, can embed and amplify discrimination at a scale that manual processes could not achieve. The Kitchener firm's algorithm did not set out to discriminate, but it may have discriminated nonetheless, and the firm will be accountable for that outcome. The experience threshold of 7 or more years, applied without flexibility and without human review, functioned as a proxy for protected characteristics, excluding an applicant whose credentials and competence were never assessed on their merits. That is the compliance gap that automated pre-hire screening must address, and understanding how it arises is essential to designing systems that avoid it.