The selection ratio is the share of applicants a hiring process actually hires. In Philippine contact-centre hiring the last figure the industry stated publicly was 6–10%, against a stretch target of 10–15%. That number is not a verdict on applicant quality. It is vacancies divided by applications, and it silently determines what your assessment can achieve, who it rejects by mistake, and how easily it fails a four-fifths check.
The ratio is tightening, not loosening
IBPAP's 2026 industry overview puts 2025 at US$40.3 billion in revenue on 1.89 million full-time employees, with 2026 forecast at US$42.3 billion on 1.94 million — revenue up about 5%, headcount up about 2.6%. In July 2026 the association revised its 2028 projection to a range of US$43.3–50.5 billion on 1.85–2.14 million FTEs, down from the 2022 roadmap's US$59 billion ambition. The downside headcount case, 1.85 million, sits below where the industry already was in 2025.
The revenue-per-head gap explains the shape of it. In 2025, contact centre and business process work ran at roughly US$20.1k revenue per FTE, against US$33.3k for IT and digital services and US$32.3k for global capability centres. IBPAP president Jack Madrid told the International IT-BPM Summit in September 2026 that the industry can no longer be measured mainly by the number of jobs it creates, with revenue projected to grow faster than employment through 2028.
For anyone running volume screening, that combination — slow seat growth, more complex work per seat, and a graduate pool that still treats the sector as a default employer — means the selection ratio gets smaller, not larger. Five beliefs that are common at a 1-in-10 hire rate do not survive contact with the evidence.
Myth 1: A low hire rate means the applicant pool is weak
The claim. "We hire 1 in 10 because 9 in 10 are not employable."
The evidence. Hire rate has demand in the numerator. If applications hold steady while net new seats grow 2.6%, the hire rate falls even if every single applicant got better. The reverse also holds: a hiring surge raises the hire rate without a single candidate improving.
The last publicly stated Philippine figures, from CCAP in 2017, described a hiring rate of 6–8% improving to 8–10%, with roughly 10 of every 100 applicants qualified and another 20–30 close enough to be worth training. Pool-level language capability has not collapsed in the years since: the Philippines scored 569 on the EF English Proficiency Index 2025, ranking 28th globally.
If you want to track pool quality, track the score distribution on a fixed, equated form over time. The hire rate tracks your requisition count.
Myth 2: At a 1-in-10 hire rate, the test barely matters
The claim. The pool is so large that anything above a rough screen will do.
The evidence. This is backwards, and it has been known to be backwards since 1939. Taylor and Russell's tables in the *Journal of Applied Psychology* showed that the practical payoff of a validity coefficient depends jointly on the selection ratio and the base rate of success without any test.
Run their model with a predictor validity of .30 and a base rate of 50%:
- Selection ratio .90 — about 52% of those hired succeed
- Selection ratio .50 — about 60% succeed
- Selection ratio .10 — about 71% succeed
Same test, same validity, triple the gain over chance. Raising validity from .30 to .40 at a .10 selection ratio lifts the figure again, to about 77%. A narrow funnel is precisely the condition under which a modest instrument earns its cost — and the condition under which a bad one does the most damage, because it is making nine rejections for every hire.
Myth 3: More screening stages means more accurate screening
The claim. Add a stage, get precision.
The evidence. Stages multiply rather than add. Four hurdles that each pass 55% of those who reach them produce an overall selection ratio of about 9% — most of your 1-in-10 may be arithmetic rather than judgement.
Worse, incremental validity only arrives when a new stage measures something the previous ones did not. Sackett and colleagues' 2022 re-analysis of selection validity put structured interviews at .42, job knowledge tests at .40, work samples at .33 and general mental ability at .31, after finding that conventional range-restriction corrections had inflated earlier estimates by .10–.20 points. Those are not figures you can simply stack.
Before adding a stage, correlate the scores from the stages you already run. If two of them correlate above .60, the second is largely buying reliability on the first construct, while adding a fresh set of false negatives that nobody ever audits.
Myth 4: The near-hire band is a training problem
The claim. Twenty to thirty of every hundred applicants nearly passed; the answer is a training programme.
The evidence. Some of those candidates did pass, and the measurement said otherwise. With a score standard deviation of 10 and a reliability of .80, the standard error of measurement is about 4.5 points, which makes a 95% band roughly ±8.8 points. Candidates sitting eight points below a cut score and eight points above it are not reliably distinguishable from each other.
The *Standards for Educational and Psychological Testing* expect test users to report the standard error of measurement and the consistency of classification decisions near a cut score, not just the pass-fail outcome. The practical consequence for a volume funnel is specific: report a band rather than a verdict, and resolve the band with a different measure — a short work sample, a scored scenario — rather than a retake of the same instrument, which mostly re-measures the same error. AssessAll reports AI-graded scenario results with integrity bands for exactly this reason: a band tells a recruiter how much of the gap between two candidates is real.
Myth 5: A standardised test cannot produce adverse impact
The claim. Everyone sits the same test under the same conditions, so the process is fair by construction.
The evidence. Fairness of administration is not the same thing as equal outcomes, and the impact ratio depends heavily on where you set the cut. Take a normally distributed predictor with a subgroup mean difference of half a standard deviation and vary nothing but the selection ratio:
- Select the top 90% — impact ratio about .87
- Select the top 50% — about .62
- Select the top 10% — about .37
The four-fifths rule in the Uniform Guidelines on Employee Selection Procedures (29 CFR §1607.4D), which the EEOC also applies to employment tests, becomes materially harder to satisfy as a funnel narrows — with the predictor completely unchanged. Roth, Bobko and Switzer's modelling of the rule in the Journal of Applied Psychology found it behaves erratically, particularly at the small sample sizes typical of a single hiring cohort.
The operational point for Philippine, Indian or any other high-volume employers is that a tightening selection ratio is itself a trigger to re-run the analysis, even though nothing about the assessment changed. Our adverse impact ratio calculator walks through the arithmetic.
What to change this quarter
- Report your selection ratio next to your hire rate, and stop reading either as a quality measure.
- Compute the standard error of measurement for every scored stage, and publish the band to recruiters alongside the score.
- Correlate your stage scores against each other before you add a sixth hurdle.
- Re-run the impact ratio whenever the funnel narrows, not only when the test changes.
- Keep the rejected-but-near-miss scores. At a 1-in-10 ratio, they are the only data that can ever tell you how many good hires the cut threw away.
The takeaway
At a 1-in-10 selection ratio, a valid assessment is worth more than it is at 1-in-2 — and a sloppy one does nine times the damage per hire. The number to govern is not the hire rate, which belongs to your requisition plan, but the width of the band around your cut score, which belongs to you.