Offer-to-join leakage is the share of accepted offers that never turn into someone actually starting work. India's PM Internship Scheme has published enough of its own funnel to measure it: across two rounds, roughly 1.53 lakh offers produced about 50,700 acceptances and around 16,000 joiners. Most of the loss happened after selection, not during it.
That is an unusual thing to be able to say with public numbers. Employers almost never publish their offer-to-join ratios, and campus teams rarely track them past the placement report. The scheme's figures — reported to Parliament, examined by a Standing Committee, and pulled apart in the press — are the closest thing India has to an open dataset on early-career hiring drop-off. They are worth reading as measurement evidence rather than as a scorecard, and they contradict four things that early-career hiring teams routinely assume.
The funnel, as reported
| Stage | Round 1 | Round 2 | |---|---|---| | Opportunities posted | ~1.27 lakh | ~1.18 lakh | | Applications received | ~6.21 lakh | ~4.55 lakh | | Unique candidates | ~1.81 lakh | ~2.14 lakh | | Offers made | 82,077 | 71,458 | | Offers accepted | 28,141 | 22,584 | | Joined | ~8,725 | ~7,300 | | Subsequently dropped out | 4,705 | 2,648 |
Offer and acceptance figures come from the Ministry of Corporate Affairs' reply to the Lok Sabha, reported in July 2025; joining, dropout and completion figures come from the Parliamentary Standing Committee report of March 2026 and RTI filings, compiled by The Wire. Dates matter here: the two sets were published nine months apart and should not be treated as a single snapshot.
One caveat before the arguments. The scheme is still formally a pilot, and pilots ramp badly for reasons that have nothing to do with selection design — delayed stipend disbursement and unfamiliar processes both feature in the reporting. Read what follows as evidence about a funnel shape, not a verdict on a programme.
Myth 1: The bottleneck is a shortage of applicants
Round 1 drew about 6.21 lakh applications against 1.27 lakh posted opportunities — roughly five applications per seat — and employers still made 82,077 offers. Supply was not the binding constraint at any point in the funnel. Whatever the shortage story says, this particular pipeline had no trouble producing candidates that partner companies were willing to make offers to.
This mirrors what is happening in the wider market. Quess Corp's FY26 Pulse report, published on 19 August 2026, found the fresher share of hiring fell from 28% in 2024 to 15% in 2025 — a demand contraction, not an applicant drought. Screening harder at the top of a funnel that is already oversupplied buys very little.
Myth 2: An accepted offer is a hire
This is the expensive one. Of about 50,700 acceptances across both rounds, roughly 16,000 people actually joined — under a third. And of Round 1's joiners, 4,705 subsequently dropped out, more than half of everyone who started.
Compound those stages and the picture inverts the usual anxiety. For every hundred offers made in Round 1, about 34 were accepted, about 11 joined, and roughly 5 completed. The screening step — the part hiring teams spend nearly all their assessment budget on — was not where the candidates went.
Most campus and volume-hiring dashboards stop at "offers rolled." If your reporting line ends there, you cannot see this failure mode at all, let alone attribute it. The first fix is not a better test; it is extending the funnel metric to joined and still there at 90 days, and holding your selection criteria accountable to that endpoint.
Myth 3: Drop-off is a stipend problem
Money is clearly part of it — the ₹5,000 monthly stipend goes further for a candidate living at home than for one relocating. But the reasons reported alongside it are structural and predictable in advance. ThePrint's reporting cites mismatch of roles, relocation and travel constraints, and the scheme's 12-month tenure — against which 73% of surveyed companies considered one to six months optimal. Attrition after joining was attributed substantially to repetitive work that did not match what candidates expected.
That is an expectations problem, and expectations are the one thing selection design can address before the offer goes out. The classic intervention is the realistic job preview — showing candidates the actual work, including its dull parts, during the hiring process. Be honest about its size: Phillips's 1998 meta-analysis in the *Academy of Management Journal* found real but modest effects on turnover, and the Quality Improvement Center for Workforce Development's umbrella review describes retention gains as small and conditional. An RJP is a cheap partial fix, not a solution.
Myth 4: Fit can't be measured before someone joins
It can be measured — but the evidence about what that measurement predicts is more interesting than the pitch usually admits. In Kristof-Brown, Zimmerman and Johnson's 2005 meta-analysis in *Personnel Psychology*, person–job fit correlated −.46 with intent to quit (k = 16, N = 3,849) but only −.08 with actual turnover (k = 8, N = 1,496). Person–organisation fit showed the same pattern: −.35 with intent to quit, −.14 with turnover.
So fit measures predict how someone feels about staying far better than whether they stay. Actual leaving is governed by alternatives, family constraints, distance and money — things a questionnaire does not control. The practical reading is not "fit measurement doesn't work." It is that fit scores belong in the conversation you have with a candidate before the offer, not in a pass/fail gate you claim will cut attrition.
What to measure instead: a five-item checklist
- Track to joined and 90-day retained, not to offer. Every downstream ratio is meaningless without this denominator.
- Assess the work, not the résumé proxy. Role-realistic scenarios do double duty: they generate a defensible score and act as a preview of the job. AssessAll's AI-graded scenario assessments are built for exactly this pairing, with integrity bands rather than pass/fail proctoring verdicts.
- Collect the logistics facts at application. Distance, relocation willingness, tenure expectation and shift tolerance explain a large share of drop-off and cost nothing to ask. Do not model them as competencies.
- Set cut scores against an outcome you actually care about. Standard-setting methods should be documented and defensible; the *Standards for Educational and Psychological Testing* and SIOP's *Principles for the Validation and Use of Personnel Selection Procedures* both expect a stated rationale, not a round number.
- Give the candidate portable evidence. Someone who declines or leaves has still been assessed. A Skill Passport that travels with them means the measurement was not wasted, and it makes a second approach cheaper.
The takeaway
India's early-career hiring problem, on the best public data available, is not that too few people apply or that too few can pass a test — it is that a third of accepted offers turn into a person at a desk, and expectations set before the offer are the lever nobody pulls. Measure to joining and to 90 days; everything upstream only earns its budget if it moves those two numbers.