More than 150,000 people submitted resumes for a coveted position in VRI’s Care Center in 2023. However, many of those applications weren’t qualified for the position, which involves answering calls and helping users troubleshoot issues.

Talent acquisition specialists Justin Syfox and Tyler White were tasked with sifting through those resumes to find the best candidates. They also had to review another 230,000 resumes for other open positions in 2023. VRI, a Modivcare service, provides remote patient monitoring to save lives and preserve independence for people in need and the caregivers who support them, so hiring top-notch support is vital.

Reviewing that staggering volume of applications was a daunting task for the two-man team. Although Syfox and White have an efficient system, screening resumes still takes about 75% of their day. That leaves less time for the next step in the applicant process—the phone interviews that give them a deeper understanding of the candidate and their fitness for the job.

“Phone screening is the most important thing that Justin and I do,” White says. “Looking at a resume, I have a good idea of who someone is. But until I have a communication with them, I’m not going to move them forward to the hiring manager.” 

Syfox estimates they spend 15 or 20 minutes talking to each applicant during the phone screening stage. In an ideal world, he’d like to spend more like 30 or even 45 minutes with each applicant. Having additional time during this step would help them ensure they’re moving the most qualified applicants forward.

So, when Slalom approached Modivcare about participating in a proof-of-concept to see if GenAI could effectively screen resumes, White and Syfox were excited to take part.

Slalom’s Troy Heitzinger shadowed Syfox and White to get an understanding of their job, processes, and requirements in a screening solution. Syfox and White provided a list of questions they consider when weeding out resumes. Slalom took that information and, using Anthropic’s Claude 2 foundation model accessed via Amazon Bedrock, prompted the GenAI model to screen applications according to role requirements provided by Syfox and White. Slalom also leveraged AWS Lambda and Simple Queue System (SQS) to build a serverless data pipeline to automate the necessary data processing, model inference, and store results.

To test the accuracy of the GenAI solution, Syfox and White gave Slalom 93 pre-screened resumes for the Care Center—46 suitable applicants and 47 unsuitable applicants. The solution re-screened the resumes by extracting data such as the applicant’s education, professional experience, and skills, then applying the screening criteria and issuing a pass or fail for each criteria. The resumes were stack-ranked based on the number of criteria passed.

In about 90% of those applications, the solution agreed with Syfox and White’s previous judgment on the resume’s overall suitability for the position. 

“I was impressed with the accuracy of the AI scores,” White says. “It was a lot higher than I expected it to be. I expected closer to 50%.” 

This successful demonstration shows GenAI’s potential to significantly reduce the workload for Syfox, White, and talent acquisition specialists across industries. Slalom estimated that on a larger scale, the solution would be capable of screening up to 20,000 resumes a day, saving each specialist 15 hours a week. This time could then be used for the all-important phone conversations, leading to more informed hiring decisions and faster job offers.

As companies in healthcare and beyond experiment with AI, they must figure out where it can best be used to make processes more efficient. This POC shows that the talent acquisition process is one area that can certainly benefit from GenAI. 

“GenAI would allow me to focus on the part of the job that is most important—getting a second impression to confirm or refute my initial impression,” White says. “At the end of the day, if we have more time for phone screenings, we get more hires through the door faster, which is what we need as a growing company.”

The POC-based approach VRI and Slalom took to deploy GenAI also points a path forward for companies still figuring out how to accelerate their deployment.

“Even though it was a relatively small sample size, I definitely believe that the results showed what we asked Slalom to prove,” White says.