by Yempo CEO Michelle Fiegehen
At Yempo, we implemented AI resume screening in 2024 to prioritise which applicants we should review first. This project came about because of the increasing use by candidates of tools that allow them to spray their CVs for a wide range of roles, including those that are far beyond their reach. These tools inundate recruiters, which in turn slows down the recruitment process giving a negative experience to both candidates and clients.
The Yempo approach to quality control
Our approach was to use the data from our 10 years of operation. Our custom-built recruitment system enabled us to build an AI model to score applicants based on previous assessments and by comparing skillsets in their CV to the minimum requirements of the role. Using this data, we can assign an applicant a number from 1-100 which allows the Talent Acquisition (TA) team to sort and prioritise which applicants will be reviewed first. Note that we do not arbitrarily eliminate anyone.
Applicants given a low rating in this system are reviewed to ensure there are no false hits, but to date it has proven highly accurate. Here are some real live examples of applicants who were given a score indicating they were unsuitable for the role:
-
- A person with 3 years of call centre experience applying for a Marketing Manager position which requires minimum 2 years in a leadership position and 5 years marketing experience;
- A person with no work experience but who had an internship as a bookkeeper, applying for a role as a Full Stack Developer which has a dozen technical skills listed as desirable, and 4 listed as mandatory;
- A person with 10 years’ experience as a registered nurse applying for a position as a Management Accountant with a mandatory CPA qualification.
What took me a while to understand was the process of building an AI model. Having an IT background, I thought we could tell the model what rules we wanted it to follow via program code, and it would follow them. For example, at Yempo we ask applicants how many years in the workforce they have, and how many different employers they have had. I wanted to set a rule to “assign a lower rating in the points scale to anyone whose number of different employers is greater than the number of years’ experience.”
This was important to me because Yempo and its clients are seeking candidates who have demonstrated their ability to commit in the long term to a role. Essentially our clients drive our principles for our service to them and therefore, our hiring principles. People who change jobs annually typically don’t have the staying power our clients are seeking, so while I don’t want them rejected – there may be valid reasons for the movements across jobs that need to be examined – I don’t want them to be considered as recommended applicants.
The challenges of AI implementation in a recruiting and outsourcing environment
Yempo’s Process Automation team set me straight. This isn’t how AI works. Certainly, a company could use automation to follow a set of rules and reject applicants based on them, which is a form of programmed discrimination, but not AI. They explained to me that our AI model was pointed at our recruitment system, the TA team, and their processes, and it monitored and then learned from their actions. We could only deprioritize applicants whose number of different employers was greater than the number of years’ experience if AI learned that this was a factor when the TA team examined applicants.
And if Yempo’s recruiters had consistently rejected anyone over 45 years old, or rejected all female applicants, AI would have learned that and stored it for future use. Clearly this is problematic. Of course, it didn’t learn these things because they are ridiculous reasons to reject an applicant and Yempo’s Talent Acquisition team doesn’t consider age or gender when it reviews their CVs. Instead, it learned exactly, organically, how Yempo processes applicants. It learned that if a candidate is earning P15k per month and their desired salary is P30k per month, they are likely not a fit for our vacancy for a senior tech specialist with a salary of P200k per month. It learned that if we are seeking an accountant with 10 years’ experience with a mandatory requirement of deep QuickBooks knowledge, and QuickBooks is not once mentioned in their CV, they are likely not a fit. These candidates are not rejected but deprioritized in the screening process. For this reason, it is vital that candidates pay attention to their CV and ensure it is accurate and answers the relevant questions and satisfies the requirements truthfully.
The Yempo AI model spent many months learning the work habits of our TA team and analysing years’ worth of data from our system, and concurrently we monitored whether the AI model would have delivered the same outcome as the team. Over time, with an accuracy rating of 99.9%, we felt confident that we could turn on the feature and start properly and fairly prioritizing candidates.
What is the benefit to Yempo and its clients in using this model?
What is the benefit to Yempo and its clients in using this model?
Since turning on our AI screening model:
-
- We have saved a considerable amount of recruiter time as they are no longer reviewing applicants who just use the scatter-gun approach to job hunting – spraying their CVs for every possible role regardless of the requirements, regardless of their qualifications, experience and expertise;
- This in turn has increased the job satisfaction of the recruiters; they spend more time chatting to qualified candidates and endorsing them to clients;
- The hiring experience has improved for qualified candidates; we are able to process them faster and get them in front of clients for interviews;
- Recruiters spend more time with qualified candidates, getting to know them and understanding their career aspirations;
- The client experience has improved because our recruiters are able to identify and endorse qualified candidates to them faster, and we don’t lose good candidates while wasting time reviewing unqualified applicants.
So who actually loses out if they are screened by an AI model that learned its rules from ethical recruitment practices?
Here are some more real-live examples from Yempo’s system:
-
- The 13,000 junior staff, mostly call centre agents, that sprayed their CV to apply for our senior vacancies even though they met none of the mandatory requirements.
- The 2,500 fresh graduates who applied for vacancies that clearly required deep levels of experience.
- That one guy who applies for every IT vacancy we’ve ever had for almost 10 years, even though he doesn’t meet any of the requirements, and who claims to have a monthly salary of P1m (USD $200k annually), far exceeding any salary we have ever offered.
I read an article that suggested listing hundreds of technologies and skills at the bottom of a CV, changing the font colour to white so it couldn’t be seen, but would still get through AI screening. This is sneaky and achieves nothing. If a candidate beats the AI screening, the next step in the process will be for a human to review the CV and if the candidate isn’t qualified, they will be rejected. Similarly we see some candidates submit applications repeatedly, changing the data in their CVs to try to get through to the next stage. This is also a dangerous practice; candidates may have their integrity questioned if a company has multiple versions of their CV on file, all with different information.
Yempo provides guidance to candidates on how they can best present themselves when applying for roles, in our blogs on How is AI affecting your job applications, Biggest CV Mistakes, Write a resume that gets noticed and Getting past the resume screening.
And finally, we can’t talk about AI screening without considering demographics. Should candidates include their date of birth, gender, ethnicity on their CV? As an Australian, I know that we usually omit these details from our CVs as they aren’t relevant to the hiring process and there are laws preventing discrimination. In the Philippines, most applicants list a full set of bio data on their CV, even including religion, hobbies, home address and a photo. I’d prefer this data not be provided but local norms prevail, and many Philippine companies require it as part of the application process.
Could it be used to screen candidates out of a role? Could they really miss out on a job because of their age or gender? Yes they could, but this can’t be attributed to AI alone. It’s either because an AI model has learnt from the TA team that this is what it should do, or the company has made a conscious decision to use automation to follow these ridiculous, unethical and discriminatory rules. Yempo does not do this. If I were an applicant and the philosophy of the company was to discriminate in this way, there is no point in trying to trick a system into progressing me to interview or hire – I ultimately will be neither happy nor welcome working in that environment.
Yempo has implemented AI in a way that replicates how our Talent Acquisition team operates, which is ethical and non-discriminatory. In doing so, it has improved the recruitment experience for qualified candidates, enriched the roles of our talent acquisition staff, and provided a streamlined and efficient service to our clients.
Thank you for visiting our blog! We hope you found our content informative and engaging. Stay connected with us for more insightful blogs, updates, and tips on outsourcing IT and outsourcing finance in the Philippines, or shoot us an email at [email protected]. We are happy to answer your questions!
