Written by Abhishek Shah
Every year, companies spend millions on unconscious bias training. And every year, the diversity numbers in their hiring pipelines barely move. HR leaders often conclude the training was not good enough, so they buy more of it. I think they are auditing the wrong thing.
Bias in hiring does not live mainly in the interview room anymore. It lives upstream, in the screen. By the time a candidate sits in front of a trained, well-intentioned interviewer, most of the biased decisions have already been made. They were made by a resume filter, a keyword match, a university name, or a recruiter skimming 400 applications in an afternoon. Training the last person in the chain does nothing about the first four steps.
Where the bias actually enters
Think about a typical funnel. A job posting attracts 500 applicants. An applicant tracking system or a tired human cuts that to 50 based on resumes. A phone screen cuts it to 12. Interviews cut it to 3. An offer goes to 1.
The interview stage, where all the bias training is aimed, handles 12 people. The resume screen handled 500. If bias removes strong candidates at the top of the funnel, no amount of interviewer awareness at the bottom can bring them back. They are simply gone, and nobody ever knows they existed.
Resumes are the problem, because a resume is not a record of skill. It is a record of access. It tells you which schools someone could afford, which companies would take a chance on them, and how well they write in a second or third language. Those signals correlate heavily with background, geography, and class. When we filter on them, we are not measuring ability. We are measuring privilege and calling it merit.
What we saw when we removed the resume
At Testlify, we run skills assessments for over 1,500 companies, which gives us an unusual view into what happens when the resume stops being the first gate. When companies put a standardized, role-relevant assessment before any human review, two things happen consistently.
First, the shortlist changes. Candidates who would have been cut at the resume stage, people from lesser-known schools, non-traditional career paths, or other countries, clear the skills bar at rates comparable to candidates with prestige credentials. The talent was always there. The paper filter was hiding it.
Second, hiring managers stop arguing about gut feel. When the first data point on every candidate is the same test, taken under the same conditions, conversations shift from “I have a good feeling about this one” to “here is how they performed.” That shift matters more for inclusion than any workshop, because gut feel is exactly where bias hides.
How to audit your own funnel
You do not need new software to start. You need three questions answered with real data from your last 20 hires.
One: at which stage did you reject the most candidates, and what evidence did you use at that stage?
If the answer is “the resume screen” and “credentials,” you have found your highest-bias, lowest-evidence decision point. That is where a work sample or skills assessment should go.
Two: does your first filter measure the job or the background?
Read your knockout criteria honestly. “Five years at a recognized company” measures background. “Can debug this broken query in 30 minutes” measures the job. Every criterion that measures background instead of the job is a place where qualified people from different starting points fall out.
Three: who never applied?
Look at your job descriptions. Long lists of requirements, degree mandates, and insider jargon quietly tell entire groups the role is not for them. Research on application behavior has shown for years that many candidates self-select out unless they meet nearly every listed requirement. Cutting your requirements list to the five things that genuinely predict success widens the top of the funnel before you spend a single dollar on outreach.
The uncomfortable conclusion
None of this means interviewer training is worthless. It means it is step four of four, and most companies have skipped steps one through three. If your funnel filters on proxies for privilege, training the interviewers is repainting a house with a cracked foundation.
The good news is that fixing the funnel is more concrete than fixing minds. You cannot easily verify whether a workshop changed how someone thinks.
You can absolutely verify whether your first hiring gate measures skill instead of pedigree, because it is written down in your process. Change the gate, and the pipeline changes with it. That is not a feel-good initiative. It is an engineering fix, and in my experience, it is the only kind that survives budget season.
Author Bio: Abhishek Shah is the Founder and CEO of Testlify, an AI-powered skills assessment platform used by more than 1,500 hiring teams worldwide.