TL;DR

Machine learning is the branch of computer science where systems improve their performance through exposure to data rather than through explicit programming. In recruitment, it's the engine behind resume screening, candidate matching, predictive analytics, and most of the AI features that ATS vendors now advertise. It's also the source of most of the bias concerns that those same vendors quietly footnote.

What Machine Learning Actually Is

Machine learning is pattern recognition at scale. A model is trained on historical data — resumes that led to hires, job descriptions that attracted qualified applicants, interview scores that correlated with tenure. It learns to recognise patterns in that data. Then it applies those patterns to new inputs: this resume looks like previous hires in this role, this candidate's profile matches the ones that converted well.

The three main types in use for recruiting: supervised learning (trained on labelled examples — hired/not hired, good performer/poor performer), unsupervised learning (finds clusters in data without labels, useful for segmenting talent pools), and reinforcement learning (learns from feedback loops, used in recommendation systems).

The critical thing to understand about ML is that it reflects whatever data it was trained on. If your historical hires were predominantly from certain universities, or skewed heavily toward one demographic for a given role, the model learns to prefer that pattern. It doesn't know the pattern is discriminatory — it just knows the pattern existed.

Why It Matters for Recruitment

Volume is the problem machine learning was built for in recruiting. A company receiving 5,000 applications for 50 roles cannot have a human read every CV. ML-based screening tools sort that volume automatically, surfacing the highest-signal candidates for human review. Whether that's a net positive depends entirely on what signal they were trained to surface.

Beyond screening, ML appears throughout the recruitment stack: in job description analysis tools that flag biased language, in chatbots that handle initial candidate engagement, in scheduling assistants that coordinate interview logistics, and in analytics platforms that predict offer acceptance probability or time-to-fill for a given role.

For in-house talent teams, the practical question isn't whether to use ML-powered tools — most ATS platforms now include some version — but how to evaluate them. What was the training data? Can the vendor explain the model's decision factors? Is there human review before any candidate is rejected? These are the questions that matter for both quality and compliance.

In Practice

A retail chain processes 18,000 applications per year for 600 seasonal roles. Their ATS includes an ML screening module trained on five years of historical hiring data. After implementation, time-to-shortlist dropped from 11 days to 2 days. Recruiter hours spent on initial CV review dropped by 70%.

In the first quarter, they also noticed their shortlists skewed 80% toward candidates from two specific postcodes — the same postcodes that dominated their historical hires, which reflected where they'd historically concentrated their recruitment advertising. The model wasn't biased in an intent sense; it was accurate to its training data, which encoded a geographic bias.

They corrected this by retraining the model after stripping postcode data from the training set and adding a post-hoc fairness audit to each shortlist. The tool remained faster than manual review; it just needed the training data corrected before it was actually useful.

Key Facts