Reejig has partnered with Holistic AI to conduct the fifth independent bias audit of our Ethical Talent AI™.
‍
This initiative aligns with Local Law #144 of the City of New York, which requires any Automated Employment Decision-making Tools (AEDT) used by employers to undergo an independent bias audit at least every 12 months. Reejig has been a trailblazer in this domain, having volunteered for the world’s first independent bias audit on a talent AI back in 2020. This partnership underscores our continued dedication to ethical AI practices and strict compliance with regulatory standards.
‍
The audit was completed on 30 November 2024.‍

Reejig is powered by the world’s first independently audited Ethical Talent AI™, which was named a 2022 Technology Pioneer by the World Economic Forum. The AI’s recommendations are based solely on skills and potential — not personal characteristics. Each customer's unique workforce intelligence delivers gender-balanced shortlists, with diversity signals on candidates to enable employers to hire, promote and mobilize talent inclusively and fairly.

Reejig enables organizations to:

• Gain 100% visibility of their workforce's skills and potential
• Reskill and upskill talent to meet business needs today, and for the future
• Curate a variety of career pathways within an organization for employees
• Understand all the work that exist within the organization and the requirements for success
• Efficiently allocate work to its workforce, including alternative workers, contractors, and alumni
• Embrace workforce automation and unlock workforce agility and optimization at scale

What kind of information is the AI trained on?

The Reejig AI model is trained using talent profile data from an employer’s internal talent systems, such as their Applicant Tracking System (ATS), Human Resources Information System (HRIS) and Learning Management System (LMS), combined with external profiles from publicly available data sources. Each employer has their own unique version of the Reejig, powered by Skills and Work Ontology tailored to their organization.

What are the characteristics and job requirements considered by the AI?

The initial step involves the creation of Longlists based on key information from job advertisements provided by the employer to refine the talent pool down to those meeting the job's essential requirements, such as location. It is important to note that the Reejig AI does not apply to the generation of Longlists.

Shortlists are then generated by the Reejig AI from each Longlist. The selection criteria used to generate the Shortlists are based on skills derived from a person’s career history, which may include the following:

• Job Experience; including where they worked, what they did in those roles, and for how long;
• Educational Experience; including where they studied, for how long, and any awards or achievements they obtained;
• Project Experience; including any projects they have led or participated in;
• Licenses and Certifications; including any relevant credentials they may hold; and
• Publications; including any books, articles or studies published by them.

How are job applicants treated by the AI?

If an individual has applied to a specific job, they will automatically appear in the Applicants section of the Shortlist, regardless of whether they were selected by the AI. The HR user may also manually add any talent profiles to the Shortlist at any time, as well as conduct further filtering or sorting of the Shortlist.

Why are the impact ratios excluded for some of the groups?

Data availability on protective attributes continues to pose a challenge as Reejig did not possess historical ethnicity data for the majority of talent profiles in the population when this audit was conducted. The Reejig AI considers both internal talent profiles sourced from the employer, and external talent profiles sourced from public data sources. The protective attributes for external talent profiles are typically unknown as they’re not part of publicly available information. The percentage of talent profiles for which ethnicity is reported adds up to less than 2% of the total population for each ethnicity group, and so those impact ratios were excluded in this audit in accordance with s 5-301(d) of Local Law #144.

The audit focuses on the selection rate of individuals from Longlists to Shortlists. Reejig provided historical data between May 14th 2022 to October 16th 2024. The data consisted of scores for 67,198 candidates in total, of which 1,199 reported both gender and ethnicity, 3,455 reported their gender alone, 58 reported their ethnicity alone, and 62,486 reported neither. The 'Shortlisted' column was considered the score column for the impact ratio calculations as this corresponds to the binary outcome of the candidate being shortlisted (1) or not (0). The 'Position ID' column was used as the deployment column and results for each are provided.