AI candidate scoring is the use of machine learning or large language models to rate a job applicant against a role's requirements, turning a resume, work history, and interview record into a numeric or ranked score. Done well, it replaces gut-feel resume review with a consistent, evidence-linked rating; done carelessly, it just moves the guesswork into a black box.
A scoring model reads the raw material on a candidate, resume, work samples, prior titles, interview transcripts, sometimes public work like code or writing, and converts it into structured signal against a defined evaluation rubric. Instead of one holistic gut call, the system rates the candidate on separate dimensions and rolls them into a score with the evidence attached to each point.
Most implementations score across a similar set of dimensions:
Candidates already know AI is in the loop. Greenhouse's 2026 Candidate AI Interview Report found that 63% of job seekers report being interviewed by AI, up 13 points in just six months, and that 70% of them were never clearly told upfront that AI would be evaluating them. For funds hiring into their own team and for the portfolio companies they back, that gap between adoption and transparency is exactly where diligence work has to focus: not whether a company uses AI to screen candidates, but whether the score it produces can be defended.
That defense matters more once a dispute happens. In Mobley v. Workday, a federal court in June 2026 limited how much of a vendor's own bias-testing data plaintiffs could compel in discovery, shielding it under attorney-client privilege. That is a reminder that a scoring system without a built-in, contemporaneous record of what evidence produced a score is a liability, not an asset, whether the dispute is a lawsuit or just a partner asking why a candidate was ranked the way they were. ScoringFactory ties every score to the line of evidence behind it for exactly this reason; see how we tie every score to a line of evidence.
The failure modes are well documented: models trained on historical hiring data can inherit the biases baked into that history, free-text resume fields let proxies for race, gender, or age leak in, and a model can be confidently wrong in ways a human reviewer would catch immediately. That is why scoring should stay human-in-the-loop, with a recruiter or hiring manager able to see and override the reasoning, and why bias mitigation has to be designed into the rubric from the start rather than patched on after an audit finds a problem.
No. It should narrow and ground the decision. The hiring manager or partner still makes the final call, but they see a ranked, evidenced shortlist instead of starting from a stack of unscreened resumes.
Keyword filtering matches text against a job description. AI candidate scoring evaluates substance, weighing experience, trajectory, and demonstrated skill against a rubric, and it attaches the reasoning to the score rather than just a pass or fail.
It should be able to. A defensible system keeps a record of the exact inputs and rubric weights behind every score, so a decision can be reconstructed months later, not just trusted at face value.
Bring a real req. We'll score the pipeline against your bar, live, with the evidence behind every number.