Machine learning is a way of building AI where software learns patterns from past examples instead of following rules a person wrote. In recruiting, most older matching, ranking and scoring tools are machine learning models trained on past hiring data.
AItopic
2 minread
2related terms
In one line
A way of building AI where software learns patterns from past examples instead of following rules a person wrote. Older matching and ranking tools in recruiting are usually machine learning models trained on past hiring data.
A rules based tool does exactly what someone told it: if the resume contains “Java”, add five points. A machine learning model is shown thousands of past examples, such as resumes and whether those people were hired, and works out its own rules. The result is a model that scores new cases based on how closely they resemble past ones.
That is both the strength and the risk. A model can spot patterns no recruiter would write down, but it also copies whatever was in the history it learned from. If past hiring favored one school, one zip code or one gender, the model can learn to favor it too, even though nobody asked it to. Large language models are a kind of machine learning too, but the term in HR software usually means the narrower prediction models behind candidate match scores, flight risk scores and job recommendations.
Why it matters to recruiters
Many sourcing, screening and ATS tools sell a match score without saying how it was built. If you know it comes from a model trained on past data, you know which questions to ask: what data trained it, how often it is retested, and whether the vendor has run a bias audit. That matters more now that several US states and cities regulate automated hiring tools.
An example from a recruiting desk
Inside one company's ATS, the in house team turns on a new match score. After a month, a recruiter notices that strong candidates from bootcamps keep landing below 40 out of 100. She asks the vendor and learns the model was trained on the company’s last five years of engineering hires, almost all with computer science degrees. The team keeps the score as one signal but stops using it to filter anyone out.
Tips that help
Ask every vendor what data trained the model and when it was last retrained.
Request the latest bias audit or adverse impact analysis before you sign.
Treat a match score as a sorting aid, never as the only reason to reject someone.
Spot check low scored candidates each month to catch patterns the model gets wrong.
Mistakes to avoid
Assuming a score is objective because a computer produced it.
Letting a model trained on another company’s hiring decide who your team sees.
Never checking whether the tool still performs after your roles or markets change.
Frequently asked questions
Is machine learning the same as AI?
No. AI is the broad goal of software doing tasks that need judgment. Machine learning is one way to get there, where the software learns from examples. Most AI you meet in recruiting tools today is built with machine learning of some kind.
Can machine learning in hiring be biased?
Yes. A model learns from past decisions, so it can repeat whatever bias was in them. That is why laws such as New York City’s rule on automated employment decision tools require bias audits, and why you should test outcomes across groups.
Do I need to understand the math to use these tools?
No. You need to know what the model predicts, what data it learned from, and how it is checked for errors and bias. Those three questions cover most of what a recruiter needs to judge a tool.
