Faster AI Hiring Still Needs Accountable Human Judgment

India’s referral practices show why processing and selection need different measures. A referral can bring a candidate to a recruiter’s attention without proving suitability for the role.

Santhanam points to the time saved in reviewing applications. “The time it takes to browse through the number of profiles, screens and shortlists is significantly reduced to minutes using AI agents and AI applications to provide the best-fit versus requirement.”

Aron expects referrals to face a more explicit assessment. “Referrals will no longer be considered just as ‘word of mouth’ endorsements; their skills and potential will be assessed by the AI system to see if they match the profile of a job.”

“Referrals may open the door, but they no longer determine who walks through it,” Singh says.

Employers can test that proposition. They can compare referred and independently sourced applicants against the same published requirements, then review who advances and why. A consistent procedure helps, but it does not prove that the requirements or their weighting are appropriate.

Language Capability Does Not Settle Cultural Interpretation

Language coverage matters in a country where employers recruit across regions, educational backgrounds, and levels of English fluency. It does not answer the harder question of how a system interprets an applicant’s response.

Santhanam is optimistic about multilingual recruitment. “AI agents will be used to identify best-fit candidates based on job requirements… HR automation will not be impacted with respect to a multi-lingual workforce.”

Aron identifies a different risk. “If a job applicant is humble about his/her qualifications, this may be construed by the AI system as a lack of confidence.”

Understanding the words in an answer and deciding what the answer reveals are separate capabilities. If English communication is essential to the job, the employer should specify the tasks and level required. If it is not, fluency needs a defensible reason to affect selection. The same discipline applies to confidence, presentation, and familiarity with formal interviews.

“Technology can capture patterns, but interpretation still benefits from human judgment,” Singh says.

Human review also needs standards. A recruiter can mistake familiarity for suitability, just as an automated assessment can reward an irrelevant signal. A human approval at the end of the process has little value if the reviewer cannot see why candidates were excluded earlier.

Santhanam cautions against assuming that more elaborate interview technology has solved this problem. “There is a lot of work that has gone into robotic interviews… with proctoring capabilities and ability to identify fake profiles,” he says. At the same time, “the reliability and integrity of such robotic interview outcomes are yet to be tested.”

Claims about fraud detection should be tested separately from claims about future performance. An eye movement, vocal pattern, or presentation style should influence selection only when the employer has job-related evidence for using it.

Santhanam describes the remaining difficulty: “Cognitive abilities, behaviours and candidate presentation… are still subject to judgement.… These elements have a cultural context… difficult to decode.”

Before adopting such a tool, employers should ask what was tested, on whom, in which languages, and against which outcome. A polished demonstration cannot establish performance across the company’s applicant pool.

Wider Campus Access Depends on Assessment Design

Online assessment can help employers reach students beyond the colleges they usually visit. Whether that broadens opportunity depends on how applicants enter the process and what the assessment rewards.

Santhanam expects automation to extend into campus selection. “Assessments of both online tests and interviews can be managed by AI agents… However… the final decision to hire or not is likely to be left to human intervention.”

Keeping the offer decision with a person leaves an earlier risk in place. A hiring manager sees only the candidates who survive screening. Someone excluded incorrectly may never reach human review.

Aron warns against reducing selection to the easiest result to count. “Campus hiring becomes a mere ‘test score affair’… The true purpose… is to find candidates who have the capacity for growth, adaptation, and fit.”

“Recruiters no longer begin with a campus; they begin with a qualified talent pool,” Singh says.

Employers therefore need to define “qualified” before a tool sorts applicants. Present ability should be distinguished from what a new hire can reasonably learn through training. “Fit” should be translated into job-related behavior, such as handling a customer complaint or working through disagreement. Left undefined, it can reward resemblance to the current workforce.

A useful review would sample rejected applications as well as successful ones. Recruiters could assess the sample with job-related work exercises and compare their reasoning with the software’s ranking. Disagreement deserves attention from both sides because a human judgment is not an unquestionable benchmark.

The recruiting team should also define a material error before rollout. Examples include excluding an applicant who meets the published requirements or assigning weight to a characteristic unrelated to the work. Applicants need a practical way to correct errors in qualifications, employment dates, or submitted documents before a position closes.