Bench hiring selects people an employer will train and then deploy across many client projects. Seat hiring selects people for one durable role inside one organisation. Bench hiring is designed to predict trainability and speed to deployment; seat hiring is designed to predict performance in a specific job. One assessment cannot optimise both, because the outcome being predicted is different.
That distinction has stopped being academic in India. The two models now sit side by side, and the numbers moved in opposite directions this year.
Two hiring models, in current numbers
The Zinnov–nasscom India GCC Landscape 2026 report counts 2,117 global capability centres in India operating 3,728 units, employing 2.36 million people and generating USD 98.4 billion in revenue — 32% growth since FY2021, on data as of March 2026. Of those, 506 belong to Forbes Global 2000 companies. The report's maturity split matters more than the headline: 13% of these centres are classified as outposts, 43% as satellites, 39% as portfolio hubs and 5% as transformation hubs.
Over the same window, the services side contracted. Business Standard reported on 23 April 2026 that the top five Indian IT firms shed a combined 6,981 in net headcount in FY26, against an increase of 12,718 the year before. TCS alone was down 23,460, and has set a fresher target of 25,000 for FY27 against its customary intake of roughly 40,000 from engineering colleges. Infosys is holding at 20,000.
This is not a story about one model beating the other. Both will keep hiring at scale. It is a story about mix: a growing share of India's hiring decisions are now seat decisions, taken by organisations whose screening toolkit — aptitude test, communication gate, high-volume funnel — was built to fill a bench.
The criterion decides the instrument, not the other way round
The *Standards for Educational and Psychological Testing* (AERA, APA and NCME, 2014) are explicit that validity attaches to a specific intended interpretation and use, not to a test in the abstract. The same logic sits at the centre of the US Uniform Guidelines on Employee Selection Procedures and of the EEOC's guidance on employment tests: a selection procedure is defensible against the job it is used for, as established by analysis of that job.
So start by writing down what each model is actually predicting.
Bench model — the criterion is training and deployment
- Completes the technical training programme
- Becomes billable inside a target window
- Can be redeployed across unrelated client contexts
- Does not exit before the training investment is recovered
Seat model — the criterion is job performance
- Performs the specific role at 12–24 months
- Owns a product area or process with limited supervision
- Communicates across time zones with an onshore stakeholder
- Exercises judgement where the right answer is not specified
Instruments line up differently against those two lists. Work-sample tests of trainability are, as the QIC-WD research summary puts it, strong predictors of success in training (Robertson and Downs, 1989) — which is exactly what the bench model needs. For job performance, the best available meta-analytic estimate for work samples is more modest: Roth, Bobko and McFarland (2005), in *Personnel Psychology*, found an observed mean correlation of .26 across 54 coefficients and 10,469 people, rising to .33 once criterion unreliability is corrected and .39 with both variables corrected. The authors note this is roughly a third below the .54 that Hunter and Hunter (1984) made famous, and conclude that work-sample validity is probably lower than practitioners assume. Structured interviews remain the strongest single method in Sackett and colleagues' 2022 re-estimation, at a mean validity of .42 with an 80% credibility interval of .18 to .66 — wide enough to be worth reading before anyone promises a number.
None of that makes a trainability test a worse instrument. It makes it an instrument answering a different question.
Four screening errors that come from applying the wrong model
1. Reusing the aptitude cut score. A score band calibrated to predict who clears a 12-week training programme carries no evidence about who will own a payments module in year two. The band may still be useful as a floor; it is not a ranking rule for the seat.
2. Assessing breadth when the job rewards depth. Bench screening deliberately avoids over-specifying, because the client project is unknown at hire. A GCC seat usually is specified — and a generic reasoning battery will not separate candidates on the domain judgement that distinguishes them.
3. Measuring knowledge where you need performance. A multiple-choice domain quiz tests recall of the domain. A seat that involves tradeoffs, incomplete information and a stakeholder on a call needs a response the candidate has to construct. This is where a scenario-based exercise with a published rubric earns its cost — and where AI-graded scenarios with integrity bands, as AssessAll runs them, make a constructed-response stage affordable at volumes that used to rule it out.
4. Scoring job-hopping instead of measuring stability. Attrition in India is genuinely falling: Aon's survey data, reported in March 2026, put overall attrition at 16.2% in 2025, down from 17.7% in 2024 and 18.7% in 2023, with nearly 75% of it voluntary against roughly 50–66% in major global markets. Industry estimates in the same report place GCC attrition lower than IT services — near 12.6% versus 13–15% — though those figures are a staffing executive's projections rather than survey output, and should be read as such. Either way, penalising a CV for short tenures is a proxy with no validity evidence behind it, in a market where short tenures were until recently the rational choice.
When bench-style screening is the right answer
Often. If the intake is genuinely high-volume and role-ambiguous — a graduate programme, a frontline ramp, an apprenticeship cohort where the first year is training — then trainability is the correct criterion and a short, well-calibrated aptitude and communication screen is the correct instrument. A three-hour simulation administered to 40,000 applicants is not defensible on cost or on candidate experience, and the 13% of GCCs the Zinnov–nasscom report classifies as outposts may not yet have roles stable enough to write a job-performance criterion for. Seat-style assessment is the right answer when the seat exists, is understood, and will still look similar in two years.
How to re-point a funnel you already run
- Write the criterion for the role in one sentence, as a 12-month outcome. If it comes out as "clears training", you are hiring for a bench, and the existing funnel is probably fine.
- Run a short job analysis against that criterion. SHRM and SIOP's competency modelling documentation is a usable template.
- Keep the cheap screen as a floor, not a rank. Floors reject; they should not order candidates.
- Add one constructed-response or work-sample stage tied to the two competencies with the largest performance spread.
- Structure the interview and score it against the same rubric. It is the highest-validity method you already own.
- Record the predictor scores and revisit them against performance ratings at 12 months. Without that, you have a process, not evidence.
For teams running both models, the cost asymmetry is the practical constraint — which is why a pay-as-you-go design, at AssessAll's ₹30 (about US$0.50) per credit, lets a seat funnel carry a deeper second stage while the volume funnel stays cheap.
Takeaway
India's hiring mix is shifting from filling benches to filling seats, and the two require assessments built against different criteria. Write down the outcome you are predicting before you choose the instrument — because the test that best predicts who finishes your training programme is not the test that best predicts who will do the job.