AI bias in hiring is real, but it's not the whole story

Plenty of HR Managers and Managing Directors have quietly decided to keep AI out of their recruitment process entirely. The reasoning is hard to fault on the surface: AI tools have been shown to encode bias, candidates deserve a human conversation, and nobody wants an employment tribunal. Research from ACCA, the global accountancy body, shows this caution is widespread. In the UK, 69% of respondents have concerns about the use of AI algorithms in hiring, and 74% of board-level leaders lack confidence in how those tools are being used. Those are striking numbers. They also tell an incomplete story.

Why UK leaders are more worried than their global peers

The ACCA findings reveal something telling: UK concern about AI in recruitment sits significantly above the global average. Globally, 48% of respondents share these concerns. In the UK, that figure is 69%. The gap is not explained by a single factor, but three themes emerge consistently from the research: potential bias in algorithms and training data, the loss of human interaction and feedback, and uncertainty over how candidate data is used and protected.

These are legitimate issues. AI tools trained on historical hiring data can reflect the patterns of whoever did the hiring in the past. If a company has historically promoted a narrow demographic, a model trained on that history may replicate those patterns. That is not a theoretical risk. It has happened at large organisations with significant technical resources. For a smaller business without a data science team to audit outputs, the concern is proportionate.

So why does the evidence tell a different story? Because the myth conflates one category of AI use with all AI use.

The myth that makes a fair point, then goes too far

The concern that AI will bake in bias is largely directed at one specific type of tool: autonomous algorithmic shortlisting, where a system scores and ranks candidates with minimal human review. That concern is well-founded. A scoring model making binary decisions about real people, without transparency or accountability, is a genuine risk.

But that is not the only way AI enters a recruitment process. Many SMEs are not evaluating whether to deploy a candidate-ranking algorithm. They are deciding whether to use AI to draft a job description, structure an interview question set, or summarise notes from a hiring panel. These are fundamentally different tasks. They carry different risk profiles. Treating them as equivalent is where the myth goes too far.

For HR Managers and Operations Managers at smaller firms, the practical effect of a blanket ban is slower hiring, inconsistent job descriptions, and more time spent on administration. Firms that have thought carefully about where AI helps and where it must not go can cut that admin without taking on the risk. The question is not whether to use AI in recruitment. It is where to draw the line.

What the evidence says about managing the risk

The ACCA research does not argue for avoiding AI in recruitment. It argues for governance, transparency, and human oversight. That distinction matters for Managing Directors and HR Managers who want to move forward without creating legal or reputational exposure.

Three evidence points are worth holding together.

First, the bias risk is real but concentrated. The ACCA research identifies bias in training data as the primary concern. This risk is highest when AI is making or heavily influencing hiring decisions autonomously. It is materially lower when AI is drafting content for a human to review and edit.

Second, UK GDPR and equality legislation already require SMEs to document how tools affect decisions, conduct impact assessments where automated processing may discriminate, and be transparent with candidates about any automated processing used. Most of these obligations already apply to any automated tool, not only AI. Knowing them gives HR Managers a clear framework for evaluating any tool.

Third, the ACCA data points to a governance gap, not a technology problem. The 74% of UK board-level leaders who lack confidence in recruitment AI are not all facing a fundamentally defective technology. Many are facing a transparency and accountability gap: they do not know what the tool does, how it was trained, or who is responsible when something goes wrong. Closing that gap through clear scoping and human oversight is a governance decision, not a technical one.

A realistic approach for SMEs

The practical path forward is not to adopt AI wholesale or to avoid it entirely. It is to scope its use precisely.

For most SMEs, low-risk applications are the right starting point. Drafting job adverts, standardising role requirements across departments, generating structured interview question sets, and reducing the administrative load on hiring managers are all tasks where AI adds value without introducing meaningful bias risk. The human still makes every decision that matters.

The higher-risk applications, automated scoring, algorithmic ranking, and CV screening without human review, warrant a much higher bar. They require vendor transparency on training data and fairness testing, documented audit trails, and a clear accountability structure for when things go wrong. Most SMEs are not ready to meet that bar, and that is a legitimate reason to avoid those tools for now.

One honest caveat: even assistive AI tools can encode assumptions. A job description drafted by an AI assistant will reflect the inputs it was given. If those inputs carry language patterns that have historically excluded certain candidates, the output may too. Human review is not optional. It is the control that makes the approach defensible.

gecco builds custom AI Assistants designed to support, not replace, human decision-making. In a hiring context, assistants built using the GRAFT methodology can help standardise job descriptions, draft interview question sets, and reduce the administrative load on hiring managers, while keeping final decisions firmly with people. No autonomous scoring of candidates, no algorithmic shortlisting.

What HR Managers and MDs should do now

For those responsible for hiring decisions, three practical steps reduce both risk and delay.

1. Map where AI is already in your process. Many businesses are using AI-assisted tools without realising it. Job board algorithms, CV parsing in ATS platforms, and scheduling tools all involve some form of automated processing. Understanding what is already there is the starting point for governance.

2. Draw a clear line between assistive and decision-making use. Assistive tasks, drafting, structuring, summarising, carry different obligations than tools that score or rank candidates. Document that distinction and make it explicit to anyone involved in hiring.

3. Check your GDPR and equality obligations before adding any new tool. If a tool involves automated processing that affects candidates, a data protection impact assessment is advisable. Being transparent with candidates about what tools are used is both a legal expectation and a trust signal.

The ACCA data shows UK leaders are more cautious about AI in recruitment than their global peers. That caution is not irrational. But caution applied without distinction between high-risk and low-risk uses means slower hiring, more administrative burden, and no competitive advantage from tools that are already being used responsibly elsewhere.

Your next step

If you are an HR Manager or Managing Director weighing up where AI can help your hiring process without creating bias or compliance risk, that is precisely what the AI Readiness survey is designed to surface. Take the AI Readiness survey. You will get access to 65+ free resources and a custom AI Readiness report. We then offer a free 30-minute showcase to walk through your results.

gecco helps UK SMEs deploy AI Assistants that keep humans in control of hiring decisions, reducing admin and standardising processes without delegating decisions to algorithms.

Website · LinkedIn · Case Studies · Newsletter