Eightfold Matching Model
Last updated: July 5, 2023
In 2021, the New York City Council enacted Local Law 144 regulating “automated employment decision tools.” In April 2023, the New York City Department of Consumer and Work Protection (DCWP) adopted rules to implement this law. Enforcement of the law began July 5, 2023.
This law includes a number of responsible AI provisions with respect to accountability, fairness, transparency, and explainability. Eightfold is providing the information on this webpage to assist the public in understanding Eightfold’s matching model that produces a match score. Eightfold is providing this information regardless of whether the model is subject to New York City Local Law 144 of 2021 or not.
Eightfold’s Matching Model
Eightfold’s cloud-based platform uses a model built using machine learning techniques. The model uses input data described below to produce a match score corresponding to a predicted degree of match between a job and a candidate, such as external job seekers or internal employees seeking promotion or other internal mobility.
The match score produced by the model ranges from 0 through 5 in increments of 0.5. The model helps predict the degree of match between a candidate and a job position to provide decision makers with objective information. It is not a stand-alone score for a candidate; rather, a match score will vary depending on the particular pairing of the candidate and the job position, as well as any changes and updates to the candidate information or the job position information.
The use of the match score can produce a list of candidates for a given job position in a rank-list manner. The match score also provides candidates with the ability to view jobs that are matched to their skills on Eightfold-powered career pages.
Eightfold provides data insights to support decision makers in Human Resource (HR) processes. The model does not replace human decision making. Regardless of the strength of a given match, candidates can apply for any open position and HR recruiting users can hire anyone they choose.
Inputs for the model
Eightfold’s model processes information obtained from both the job position and the candidate. The model works with job related information such as titles, work experiences, education, skills, and natural language of resumes and job descriptions. The job position information may include the employer’s job descriptions, job postings, or inputs provided by an employer representative such as desired skills, experience, or education. The candidate typically provides a resume or a job application that conveys work history, education, skills, and so on.
Viewing match score outputs
Using the data inputs described above, the model produces a match score between a given job and a given candidate. Eightfold’s platform can display to employer users the match score of the position-candidate pair and can also provide explanations on how the model derived the score. These explanations include concise breakdowns on how well a candidate profile matches the requirements of the job, including for example skill relevance, work relevance, and job title relevance.
When reviewing each position and candidates for the position, the employer representative can use the model output to help understand the specific skills, titles, work, and job title relevance of each candidate for the position. The reviewer can hover over a given candidate to see the breakdowns and can also click to view the candidate’s work experiences. The reviewer can view candidates in any order and can choose actions regardless of the display of the model output.
When visiting an Eightfold-powered career site, a candidate can upload a resume and view jobs that may be a good match. The display is designed to surface more opportunities to the candidate based on objective information. The candidate may choose to apply or not apply to any of the jobs surfaced and may search for additional ones to examine.
Eightfold has completed a third-party bias audit of the Eightfold Matching Model. The date of the most recent bias audit is June 16, 2023.