Jobvite is committed to educating our customers, prospects, applicants, candidates, and the general market about our efforts in the artificial intelligence (AI) and machine learning (ML) space. The industry is moving in the direction of more automation driven by AI and ML, resulting in increased activity that is guided and/or executed based on pre-defined workflows and data models. This advancement will provide amazing productivity increases, allowing customers to create and build relationships with many more individuals in each phase of the sourcing, recruiting, hiring, and onboarding journey. Jobvite will still use the same communication mechanisms employed today, such as phone calls, email and text messages, and video conferencing, but much of the communication, screening, assessments and rankings will be powered by processes and data defined to maximize velocity, reduce time-to-fill, and improve the overall experience for both candidates and recruiters. To ensure effectiveness, validity, and fairness of these automated processes and AI/ML data models, it is critical to provide clarity to all individuals involved, so that we understand how and why decisions are being made. Jobvite is committed to exposing the logic behind these processes, for both explainability and accountability.
All parties have a right to know how their data is being used and interpreted. The descriptions below detail how Jobvite uses AI/ML algorithms within our platform. This document will be updated as new algorithms and data models are introduced.
- Jobvite evaluates each candidate at every step of the process. We review all available, non-personally identifiable information (PII) data points, including but not limited to resume, job description, conversations, messages, applications, and assessments. Each piece of information is consumed and processed so that Jobvite can quickly identify and highlight the individuals that most closely match the preferences of the hiring team.
- Data that is specifically removed, if presented, from consideration of the model.
- Email address
- Phone number
- Specifically, Jobvite evaluates a candidate's match to a job description based on the candidate’s skills and previous jobs found in the candidate’s resume.
- Each resume is parsed and a list of skills are extracted.
- Jobvite uses the skills list to create a skills database that links skills of an applicant to an anonymous ID.
- The anonymized skill data is aggregated and processed via AI to identify and create clusters of skills that are similar. Skill similarity is determined by analyzing the correlations and overlap of skills that exist on resumes.
- Using the skills parsed from the resume as a filter, the job description is parsed and skills are extracted and processed in a similar way.
- Jobvite uses the skills database to determine which skills are the same, which are similar, which can reasonably be inferred, although not expressly stated, and which are dissimilar, creating a layer of interpretation utilized by the algorithm.
- This interpretation per skill from the aggregate data is then applied to each resume and job description in the database, in an identifiable way.
- Leveraging the covariance structures of the job description and resume skills, Jobvite predicts the likelihood of a candidate possessing skills identified in the job description given the skills on their resume. Probabilities are transformed into grades which are displayed to the user.
- When a new job description is posted and a candidate match is requested, Jobvite matches candidates based on the similarity of the skills in the job description to the skills in a candidate’s resume.
- Job trajectory
- Each resume is parsed and up to the last seven job titles are extracted from the applicant resume. No times or durations of employment are extracted from a job title.
- Jobvite uses the job title words to create a job title database that links words in an applicant’s job history to an anonymous ID.
- Jobvite uses the job title database to learn the relationships between job titles and how job titles progress and change over a career. Jobvite achieves this objective by creating a large matrix that identifies relationships between one job title at time t and the next occurring job title at time t+1. For example, the probability of an applicant having the next job title “Senior Engineer” given their current job title is “Sous-Chef” is less likely than that of an applicant with a current job title of “Engineer”.
- Probabilities are transformed into grades which are displayed to the user.
- When a new job description is posted and a candidate match is requested, Jobvite matches candidates based on the likelihood of the words in the applied to job title being the words in a candidate’s next job title as predicted by the job history on the candidate’s resume.