What Unilever’s AI Hiring System Looks Like Now

A recent graduate sits in an apartment looking into a laptop camera. There is no hiring manager, no interview panel, only timed questions shown on the screen. In the background a computer program decides if anyone at Unilever will ever see her application.

Unilever has used this kind of process since 2016. Each year the company hires about 30,000 people from 1.8 million applications and, for entry-level jobs, it redesigned the system so that software does most of the screening before any person gets involved.

What Unilever wanted to achieve with AI

The main challenge was early-career hiring. Its Future Leaders Programme received 250,000 applications for about 800 roles in over 50 countries. Recruiters couldn’t review every application and the shortcuts they used, like focusing on certain schools and keyword filters, made the selection group keep favoring the same types of backgrounds.

Unilever wanted to cut the hiring time from months to weeks, reduce manual screening and increase diversity without hurting quality. The first setup for this work used Pymetrics (now under Harver) for early tests, HireVue for AI-scored video interviews and partners such as Amberjack to design and run the process.

The operation of the process, in 2025

Now most applicants start the process online. They find a job-board or LinkedIn post, fill out an online application and go straight into the assessment process instead of waiting in line at a campus stand.

The first real filter is a set of game-like tests. Taken from the Pymetrics library and now offered by Harver, these ask candidates to do tasks that measure focus, memory, comfort with risk and how they make decisions. In the background the platform compares these patterns with profiles built from Unilever’s strong performers in similar jobs and creates a match score.

This is where survivorship bias shows up. The model learns only from people Unilever has already hired and kept; it never sees the candidates the old process turned away. It focuses on "who has done well before" rather than "who might do well if we changed how we hire."

Applicants who pass the cutoff move to an on-demand HireVue interview. They answer set questions using their phone or laptop. The system now scores both the transcript and the audio, looking at what they say, how clearly they speak and how well their answers match the skills Unilever wants for the role.

A small top-ranked group moves on to the "Discovery Centre" assessment days. There, applicants do group tasks and case studies while hiring managers and HR decide the results. The AI decides who gets into the room; it does not decide who gets the offer.

Before and after: what really changed

Before Unilever redesigned this process, the graduate hiring cycle lasted more than four months. Recruiters spent weeks on phone screenings and manual CV checks, and some candidates dropped out at different points between the first contact and the final interview.

After the AI rollout Unilever says time-to-hire for these roles fell to four weeks, cutting the process length by almost 90%. It also reports saving 50,000 hours of interview time in the first 18 months and over £1 million a year in hiring costs by removing travel, manual screening and extra interviews.

On the candidate side, completion rates at each step went up to about 96%, compared to almost half of applicants dropping out before, and Unilever links the AI-led process to about a 16% increase in diversity among early-career hires.

The difference before and after is clear: timelines cut from months to weeks; heavy manual screening replaced by AI sorting; more people getting through the process; and a more diverse group at the end.

Did the hires actually improve?

Speed on its own is not enough. If faster, cheaper hiring leads to weaker teams, the case study falls apart.

Unilever’s internal data shows a pattern. From reviews of the program the company says that for candidates reaching the final stage, offer rates rose from 63% to 80% and acceptance rates went from 64% to 82% after the new process. This suggests a better match: fewer wasted final rounds and more candidates accepting roles that fit them.

Later write-ups describe the system as a way to "scale hiring efficiently while improving candidate quality and diversity." The fact that Unilever has kept this setup for years is a strong sign. What you don’t see is any public number for first-year retention or performance, so making stronger claims would go beyond the data we have.

The facial-analysis detour

Part of the story is often missed. Older versions of the HireVue system looked at facial expressions and micro-expressions as well as voice and words. Civil liberties groups and researchers questioned how valid and fair this was, and the Electronic Privacy Information Center filed a complaint with the US Federal Trade Commission saying the system was unclear, invasive and biased.

In 2021 HireVue said it would stop using facial analysis in candidate scoring. The current Unilever process does not use this feature.

The lesson for AI hiring

After eight years Unilever's hiring process shows both the strengths and limits of AI in recruitment. When used carefully, automated screening can cut time-to-hire and costs, keep more candidates in the process and improve diversity, as long as the process is well managed and the results are tracked.

The hidden trap is in the data and the signals you choose. If you train only on yesterday’s "top performers," survivorship bias appears by default. Add signals like facial analysis and you invite questions from regulators.

If you copy anything from Unilever’s case, copy the structure, not the list of tools: let AI handle the repetitive pattern-matching at the top of the process, keep people in charge of the final decision, and treat your training data and feature choices as live risk decisions you review often, not assumptions you leave buried in the fine print.