Talent Fit
Employ’s Talent Fit tool is available in JazzHR, Lever, and Jobvite. Candidates are evaluated automatically upon application. Candidates are re-evaluated in the event that the Job Description is updated in the ATS. Talent Fit uses only the Job Description and the candidate’s resume to determine if the candidate is a fit for a job. No other data is included from the candidate’s application or other job details.
Talent Fit Process
First, the candidate's resume is anonymized, and then compared to the job description by prompting a large language model (LLM) to evaluate the candidate’s fit based on the identified skills and experience in the job description. The LLM is instructed to act as a recruiter and evaluate a candidate without regard to age, race, gender, or disability.
Talent Fit discerns what is relevant and less relevant information by directly using the job description in determining whether the candidate is a good fit for the role. The tool only evaluates education, skills, and experience that are called out as necessary by the job description. Candidates are holistically assessed – there is no formula or set of ‘boxes’ the candidate needs to tick to be considered a ‘fit’ for the job.
Results
We provide a binary decision on whether the candidate is a suitable fit, along with an explanation that includes the candidate’s strengths, key considerations, and areas for clarification. Areas to clarify generate a list that the recruiter might use to get more information if there is any uncertainty. For example, a Construction Project Manager candidate may not have listed the size of the project they manage, or a School Bus Driver candidate may not have listed the type of driver’s license that they have.
No job is guaranteed to have a certain number of matches, and each candidate is evaluated individually, with no consideration of the other candidates in the pipeline.
Talent Fit Technology
We utilize both AWS Bedrock and IBM’s WatsonX governance to facilitate our AI functions, in conjunction with third-party LLMs. These vendors are considered subprocessors of Employ.
Our Talent Fit feature utilizes external, closed-source LLMs (Large Language Models). We don't manage or maintain our own models, and we don’t conduct any model training on hiring outcomes. We directly compare each resume to the relevant Job Description to evaluate if the candidate is a strong fit for the job.
Employ reserves the right to determine which LLMs / GenAI providers we will use. Due to the rapid pace of change in AI, we do not send notifications to customers when we transition from one LLM to another, aiming to maximize accuracy and balance costs, unless we deem it relevant to the customer.
We maintain a testing dataset that we use for accuracy verification. If that accuracy decreases at any point, we reevaluate the prompt and LLM being used to achieve a high level of accuracy consistently. Due to the testing framework we have built, this can be done in a matter of hours, allowing us to detect a drop in accuracy. We also utilize this testing framework to test the latest models and methods as they are made available.
Talent Fit Data Security
Data is anonymized before leaving Employ’s environment, and because we don’t do any model training, data is never consolidated across customer environments.
Preventing and Mitigating Bias in Results
During the anonymization process, we remove any identifying information, including data that may reveal race, age, gender, or disability status. In addition, the Talent Fit prompt specifically instructs the LLM to disregard any data that alludes to these, in the event that the anonymization were to miss any data that indicates race, age, gender, disability, or veteran status.
IBM watsonx.governance tracks Employ’s compliance as a software provider, ensuring for our clients that the AI functions we create are free from bias. When evaluating bias in AI, we use a method similar to the way the EEOC determines if a hiring practice has resulted in an adverse impact. They use an impact ratio of 0.8 when measuring the selection rate of the minority group versus the selection rate of the majority group. We will hold ourselves to a higher standard, ensuring we meet an impact ratio of over 0.9. Should our number dip below that, we will be alerted immediately by IBM watsonx.governance and rectify immediately.
Employ will proactively share compliance updates related to our AI tooling. These updates will be distributed broadly across our customer base.
Notifying Customers and Candidates of AI Use
Our customers must turn on Talent Fit – it is not automatically enabled in any customer environment. By turning it on, the customer consents to using AI in their hiring process.
Customers bear the responsibility of informing candidates that AI is being used to evaluate their applications. This is often done in the Job Description itself or can be done on the career site.