When AI Screens AI: How the Tech Hiring Funnel Is Breaking Down From Both Ends
A hiring manager at a mid-size technology company described their experience last quarter: they posted a senior cloud architect role, received 620 applications in 72 hours, and their ATS shortlisted 22 for human review. They interviewed 8. They hired nobody. The quality of all 22 was not what they needed, despite the volume that came in.
This is the current state of tech hiring at scale, and it is getting worse. The system is generating more volume and delivering less signal at every stage.
How We Got Here
The dynamic is structurally straightforward. AI tools lowered the cost of applying to near zero for candidates, which increased application volume. That volume made it impossible for TA teams to review applications manually, which accelerated the adoption of AI screening tools. Those AI screening tools are now filtering AI-generated applications based on AI-scored criteria. The humans in the middle are either rubber-stamping what the algorithm recommends or spending their time on escalations rather than on judgement-based assessment.
According to a 2025 report from the Harvard Business School Managing the Future of Work programme, AI-screening tools now reject approximately 75 percent of applications before any human reviews them. A separate analysis from recruiting technology firm Greenhouse found that the average enterprise tech role now receives 4.3 times as many applications as it did in 2021, while the number of interviews per hire has increased by 34 percent over the same period, suggesting that the shortlisting quality has declined even as volume has increased.
What the Algorithms Are Losing
ATS keyword-matching and AI scoring tools are calibrated to historical data about what a hire who progressed through the funnel looked like. They are not calibrated to what the business actually needs from this specific role, because that information is rarely surfaced in a form the algorithm can use. The result is systematic filtering for proxies, degree from a target institution, keyword match to the JD, gap-free employment history, that have weak causal relationships to job performance.
The candidates who lose in this environment are not the weakest candidates. They are often the candidates whose backgrounds do not fit the algorithmic template: the engineer with a non-linear career path that included a startup, the domain expert from a sector adjacent to the one the JD specifies, the professional who took a parental leave gap that the algorithm interprets as a red flag. These are exactly the candidates who often bring the differentiated perspective that the hiring organisation says it wants but its screening system filters out.
Research published in the MIT Sloan Management Review found that algorithmic hiring tools systematically disadvantage candidates from non-traditional backgrounds by 35 to 40 percent compared to human reviewers making the same assessment. The irony is that organisations paying for AI screening believe they are reducing bias. In practice, they are encoding historical bias at scale.
What the Data Shows About Quality
The quality-to-shortlist ratio, the proportion of submitted profiles that progress to interview, is one of the clearest measures of funnel efficiency. For inbound applications managed through standard ATS systems, this ratio typically runs between 5 and 15 percent. For sourced shortlists assembled by a specialist recruiter who has conducted a genuine role intake conversation, the ratio runs between 40 and 70 percent.
The difference is not the AI. The difference is whether the search started with a clear, specific brief about what success in this role actually requires. AI screening cannot compensate for a brief that was never written. It can only filter for signals that correlate with past patterns, and past patterns include all the historical misfires that the organisation has made.
The Actual Fix Is Not More AI
The organisations producing the best quality-to-shortlist ratios in the current market are not the ones with the most sophisticated ATS deployments. They are the ones that have invested in what happens before the funnel opens: a genuine intake conversation with the hiring manager, a clear and specific brief that goes beyond the JD, and a sourcing strategy that targets the right people rather than accepting whoever opts in.
That means resisting the temptation to add another AI layer downstream when the real problem is upstream. It means TA leaders who have the standing to push back on hiring managers who want to post a JD and wait for volume. And it means measuring shortlist quality rather than application volume, because the metric you track determines the system you build.
The AI arms race in hiring is a symptom of an intake process that was never properly designed. Fix the brief, and the screening problem largely solves itself. Add more screening AI without fixing the brief, and you get faster processing of a fundamentally broken signal.