United KingdomConsumer goods2017 – ongoingUnilever
AI screening of early-career applicants with Pymetrics and HireVue
Unilever screened early-career applicants with games and AI-scored video interviews, saving 50,000 hours by the vendor's count
Problem
Unilever received up to 250,000 applications a year for early-career roles. Sifting them by hand took months and a large share of recruiters' time.
Approach
Applicants first played online assessment games from Pymetrics, then recorded answers to set questions in a HireVue video interview that software scored, at the time including facial expressions, body language and keywords. Those who passed were invited to an in-person Discovery Center stage, from which about 800 received offers. In January 2021 HireVue announced it had stopped using facial analysis in its assessments, saying language analysis made the visual signals add little, after public criticism and a 2019 complaint to the US Federal Trade Commission.
Results
Takeaway
The vendor later dropped facial analysis from applicant scoring, which carried fairness and transparency risks. Time and cost savings were vendor claims without independent verification.
Categories
- Size
- Large enterprise
- Industry
- Consumer goods
- Business function
- Human resources
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and MLComputer vision
- Adoption
- Vendor solution
- Evidence
- Vendor's case study
Sources
Source summaries
Best Practice AI reported HireVue's account of hiring at Unilever. HireVue said it had taken four to six months to sift 250,000 applications for about 800 roles, and its video analysis of facial expressions, body language and keywords saved over 50,000 hours of candidate time in 18 months and more than £1 million a year. HireVue also claimed a 90% shorter time to hire and 16% more diverse hires.
The German business magazine describes Unilever's process of up to 250,000 applications a year screened through Pymetrics games and recorded video interviews. It says automated screening saved about 75% of recruiters' hours spent interviewing and assessing candidates, and that about 3,500 candidates reach an in-person Discovery Center stage, from which about 800 receive offers.
Fortune reported that HireVue's algorithms no longer considered candidates' facial expressions, a change the company made partly in response to criticism including a 2019 complaint to the Federal Trade Commission by the Electronic Privacy Information Center. It also covers an outside audit of HireVue's assessments, which raised questions about how the system scores candidates with different accents and very short answers.
The Society for Human Resource Management reports that HireVue announced in January 2021 that it had dropped facial analysis from its screening assessments, saying the feature had been discontinued in March 2020. The company said improved language analysis meant visual analysis no longer added much value, and its chief executive said the visual component was not worth the concern it caused.
Full details are in the original sources.
