High-volume hiring breaks on two chores: reading hundreds of resumes against the spec, and the scheduling back-and-forth that eats a recruiter's day. Automation does both well — rank every applicant against the role with a written reason, and coordinate interview scheduling without the email tennis — provided one line holds: the agent shortlists and schedules; it never auto-rejects a candidate with no human in the loop. Screening is where bias and legal exposure live, so the decision stays with a person. Here is how the two workflows run, what the public numbers really show, and where it is the wrong build.
The two workflows
Screening: every inbound resume read against the role spec you wrote down, ranked, with a plain-language reason for each placement, and a shortlist handed to a recruiter. The strong candidate sitting in the back of a pile of three hundred still surfaces, and every ranking is logged so a decision can be explained after the fact.
Scheduling: self-serve slot booking, reminders, and reschedules, so the recruiter stops playing calendar tennis across dozens of candidates. Both workflows remove volume drudgery; neither makes the hire. That division is the whole design, not a caution bolted on.
The evidence, read honestly
Paradox publicly reports that for 7-Eleven, conversational automation of applicant scheduling and coordination got 85% of applicants scheduled within an hour and saved store leaders around 40,000 hours a week. HireVue publicly reports that for Great Southern Bank, CV screening plus asynchronous video assessment cut time-to-hire from more than 40 days to 23, reduced screening time by about 60%, and saw 38% of assessments completed outside business hours. Those are the vendors' reported results for their clients — third-party public evidence from US and Australian high-volume contexts, not Orkivanta benchmarks.
The transferable point is not the percentages; it is which chores moved. Scheduling coordination and first-pass screening are the repetitive, high-volume work automation is strong at — and in both cases the hire itself stayed with a human. Borrow that shape. Leave the numbers, which are bound to those companies' baselines and roles.
The India layer, and the line that does not move
The volume roles here are retail, BPO, delivery, and field sales, with multilingual applicants and WhatsApp as the channel candidates actually answer for scheduling. At volume, the spec discipline matters more, not less: separate the must-haves from the nice-to-haves before any screening runs, because a quiet disqualifier does not reject one candidate, it rejects hundreds. The ranking should ground in that explicit written spec, with an auditable reason per placement (the pattern behind our Recruitment Resume Screening Agent, at /products/resume-screening-agent), not in a hidden model of a 'good candidate.'
The line that does not move is the no-auto-reject rule. The agent ranks, explains, and schedules; a person reads the shortlist and decides. A screener that rejects candidates to zero human involvement is the exact thing that turns a hiring process into a legal problem, and grounding the ranking in a written spec rather than in 'who we hired before' is the defence against automating yesterday's bias. Fairness here is a human-review requirement, not a certification a vendor can hand you.
When it is the wrong build
When the roles are low-volume and you already read every applicant carefully — there is no drudgery to remove, so there is no payoff.
When the spec is vague. If you cannot say what the role actually needs, the agent has nothing real to rank against and falls back on inventing criteria — the failure you were trying to avoid. Write the spec first; that work is not the tool's to do.
And when the team will not review the shortlist. The no-auto-reject rule only protects anyone if a human actually reads the shortlist and the reasons. If nobody does, you have built an auto-rejecter with extra steps and a false sense of safety. The safeguard is a person showing up.