No topic distorts India-entry decisions like attrition. Boards carry a number in their heads — usually 25%, usually from a 2022 headline — and discount every business case accordingly. The actual data tells a different, more useful story: attrition in Indian tech is cyclical, segmented, and above all managerial. This analysis reconstructs the real numbers from public disclosures, reviews what exit research says about causes, and ranks the interventions by evidence rather than fashion.
The idea in brief. The 25% horror numbers were real — for services companies, at the peak of the 2021–22 frenzy, visible in listed firms’ own investor disclosures. The durable baseline for a competently run GCC is 10–15% voluntary attrition, comparable to tech hubs worldwide, with well-run centres posting single-digit regretted attrition in the same cities where weak ones bleed a quarter of their staff. Exit research consistently ranks growth stalls and manager quality above compensation as causes. Attrition is not weather; it is feedback — and the interventions that move it have measurable returns.
Reconstructing the real numbers
The full cycle is visible in one chart, built from the quarterly disclosures of India’s largest listed IT-services companies and industry reporting on captive centres:
Three readings matter:
- The spike was real. In FY2022, the majors reported quarterly attrition annualising above 20–25% — check the investor-relations archives of TCS and Infosys and you will find the numbers in their own slides. Demand had exploded, salaries repriced quarterly, and counter-offer culture reached its historic peak. Every scare statistic since descends from those four quarters.
- The cooling was equally real. By FY2024 the same disclosures showed attrition back in the low-to-mid teens — a normalisation reported across the industry as hiring rationalised.
- GCCs run structurally below services. Captive centres typically report several points under the services industry — better pay bands, single-employer mission, product work — and the differential persists across cycles, which is why the teal band sits below the gold line throughout.
The planning conclusion: model 10–15% voluntary attrition for a well-run centre — hot-skill segments (AI/ML, data platform, security) above the band; domain-operations roles and calmer cities like Pune below it. Anyone still modelling 25% is pricing 2022’s panic into 2026’s plan; anyone modelling 5% is pricing a fantasy. And recall from our cost analysis what the band is worth: the difference between 10% and 25% is roughly 38 index points of cost per retained employee — the entire India advantage at the margin.
What the research says about why people leave
Exit-survey aggregates across the industry, echoed by the broader organisational literature, rank the drivers with notable consistency:
- Growth stalls (the dominant driver). No visible next role, no new skills, no forward motion. India’s workforce is young and ambitious; in a market with 1,700 alternative employers, stagnation reads as a signal to move. Career-path opacity is a resignation letter in draft.
- Manager quality. The global truth — Gallup’s State of the Global Workplace series attributes the large majority of engagement variance to the manager — with local amplification: India’s rapid growth promotes first-time managers faster than it trains them. The untrained layer becomes the silent attrition engine, invisible in dashboards because its effects scatter across “better opportunity” exit codes.
- Work that bores. Centres holding only maintenance mandates lose precisely the people capable of more — adverse selection in its purest workplace form. The mandate your centre head negotiates from HQ is, quite literally, a retention instrument.
- Compensation — real, but later than assumed. Pay triggers exits mainly when one of the above is already broken. The fairly paid, growing, well-led engineer rarely moves for 10%; the stalled one moves for less. Treating attrition as a pricing problem alone is the costliest mis-diagnosis in the field.
The interventions, ranked by evidence
| Intervention | Mechanism | Evidence strength |
|---|---|---|
| Visible internal promotion & published career ladders | Attacks driver #1 directly | Strong — cheapest retention lever available |
| First-line manager training & coaching | Attacks driver #2 | Strong — among the best-ROI people investments measured |
| Mandate expansion (owned products, AI initiatives) | Attacks driver #3; retains the best specifically | Strong, though slower to arrange |
| First-year design: onboarding, buddy, month-6 stay conversations | Attrition concentrates in months 3–12; design absorbs it | Strong — see the ramp economics in our cost analysis |
| Proactive band correction against live benchmarks | Removes the pay trigger before it fires | Moderate-strong; timing is the skill |
| Reactive counter-offer bidding | Delays exits it does not prevent | Weak — a large share of accepters leave within a year regardless |
| Perks and engagement theatre | Addresses no ranked driver | Negligible where the top three are broken |
Two of these deserve elaboration. Manager training is the least glamorous line and the most reliable: a first-time manager coached in feedback, career conversations and workload design retains a team; an uncoached one quietly seeds next quarter’s exits. Stay conversations at month six exploit the timing science — attrition risk peaks in the first year — and cost nothing but calendar discipline. Both are fully within a centre’s control, which is the point: the highest-evidence interventions require no HQ approval and no budget cycle.
The counter-offer trap, briefly
India’s long notice periods (30–90 days) give employers weeks to woo back a resigning employee, and a counter-offer culture has industrialised around that window. The data is blunt: counter-offer acceptance mostly defers departure — a large share leave within the year, having meanwhile learned that resignation is the fastest promotion channel, a lesson the rest of the team observes carefully. Budget for the proactive interventions above; treat reactive bidding as the tax on having skipped them. (The offer-to-joining version of this dance — where the same window attacks your incoming hires — gets its own full analysis in article 28.)
Reading attrition as information
The reframe that separates sophisticated operators: attrition is a lagging indicator of management quality, segmentable into signal. Regretted vs non-regretted — losing your best is a crisis; losing your bottom decile is often the system working. Cohort timing — month-3 exits indict onboarding; month-18 exits indict career paths. Team clustering — three exits under one manager is not a market trend; it is a manager. Centres that run this diagnosis quarterly treat retention as an engineering discipline, and their single-digit regretted-attrition results — achieved in the same cities, competing for the same talent, as centres losing 25% — are the strongest evidence in this entire debate that the variable is management, not geography.
A 90-day retention agenda
- Instrument the diagnosis: regretted flag, cohort curves, manager-level clustering. One blended number cannot be managed.
- Publish career ladders and run the next promotion cycle visibly.
- Put every first-time manager through structured training; coach the struggling ones now.
- Schedule month-6 stay conversations for the current joining cohort.
- Re-benchmark hot-skill bands against live data; correct proactively, once, rather than reactively, repeatedly.
The segment gallery: attrition by role family
Centre-level averages conceal the pattern that planning actually needs. The role-family gallery, as indicative bands consistent across industry reporting and engagement experience:
- Hot-skill engineering (AI/ML, data platform, cloud security): the top of the range — mid-teens to twenties even at good employers — because every employer in the country is bidding (the domain analyses in articles 11–13 map the demand side). Strategy: relative retention, deep benches, and growing successors in parallel rather than pretending the tax away.
- Mainstream software engineering: the centre’s average, roughly tracking the 10–15% planning band; most responsive to the growth-and-manager interventions, which is where the leverage lives.
- Domain operations (finance, healthcare, compliance): below average — domain professionals change employers more deliberately, and fewer rivals bid (the F&A and pharma analyses document the loyalty differential). These families are the retention ballast of a mixed centre.
- Support and quality functions: historically higher-churn in services settings, dramatically stabilisable in captives via the earn-scope paths that services vendors structurally cannot offer — among the clearest captive-model dividends.
- Early-career cohorts: bimodal — near-total retention where graduate programs are real (mentors, rotations, visible ladders), and the industry’s worst numbers where freshers are benched or drifted. Cohort design, not cohort quality, decides which mode you get (article 21 carries the design).
- Leadership and anchors: low frequency, catastrophic amplitude — a single anchor exit can cascade (the mis-hire economics in article 6). Managed individually: annual package benchmarking, mandate reviews, succession depth.
The planning use: set family-level targets and family-level interventions, then let the centre average fall out — the reverse of the common practice, and the version that survives contact with a mixed 200-person reality.
A worked retention P&L
To convert the argument into budget language, model a 200-person centre at the two ends of the management-quality spectrum — composite figures, transparent arithmetic:
- At 20% voluntary attrition (weak version): 40 exits a year. Each costs, conservatively, a recruiting fee, three months of vacancy and three months of successor ramp — call it 60–80% of an annual salary all-in. The annual attrition bill approximates 25–30 full-time salaries — invisible on any single line, paid in full every year.
- At 11% attrition (designed version): 22 exits, roughly 14–16 salary-equivalents — a recurring saving of 10–15 salaries a year.
- The intervention budget that buys the difference — manager training for every first-line lead, a real onboarding program, stay-conversation discipline, proactive band corrections, career-ladder administration — prices, generously, at 3–5 salary-equivalents annually.
A three-to-five-fold recurring return, before counting what the model cannot see: unshipped features, HQ trust erosion, the compounding referral networks that leavers take with them and stayers build. This is the calculation that reframes “soft HR investments” as the highest-yield line in the centre’s budget — and it is robust to halving every assumption in it.
Questions leaders ask about attrition
“What number should I promise my board?” Promise the diagnosis, not a number: voluntary attrition of 10–15% at steady state, single-digit regretted attrition as the management target, and quarterly reporting on the segmented dashboard (regretted/cohort/cluster). Boards accept honest ranges; they punish confident fictions — usually eighteen months later.
“Is high attrition ever acceptable?” Yes, in two honest cases: deliberate churn at the performance floor (the system working), and hot-skill segments where the market sets a tax no employer escapes — there you compete on relative retention and bench depth, not absolutes. What is never acceptable is not knowing which case you are in.
“Do retention bonuses work?” As a bridge across a known cliff (post-acquisition year, project completion), modestly. As a standing strategy, no — they convert engagement problems into deferred-compensation problems and teach the team that leaving-threats price better than staying. The interventions table above orders the alternatives by evidence.
“How is AI changing attrition?” Two visible effects: task-shaped roles carry rising anxiety (address with visible reskilling paths — see our AI analysis), while AI-skilled staff face intensified external demand (address with genuine AI mandates, not retention lectures). Both reduce to the same underlying driver — growth — that has topped the exit research for a decade.
“We’re at 25% now. Where do we start?” Triage order: segment the data this week (the three-way diagnosis costs nothing); interview the last ten regretted leavers’ managers and teams; fix the two or three manager-clusters the data will almost certainly reveal; then build the systematic machinery. Centres have moved from 25% to low teens inside eighteen months on exactly this sequence — the levers are strong precisely because the baseline was neglect.
The brand feedback loop: attrition as public information
One dynamic completes the systems view: in India’s reputation-driven market, your attrition is not private. Employer-review platforms, alumni networks and the referral grapevine publish it — informally, continuously and with commentary. This creates a feedback loop that operates in both directions:
- The doom loop: visible churn → wary candidates → weaker funnels → panic hiring at premiums → quality dips → more churn. Centres inside this loop find every hire harder and pricier than the market average, and often misread the cause as “market conditions.”
- The flywheel: visible stability → alumni advocacy → referral-rich funnels → better selection at calmer prices → stronger teams → more stability. The best centres recruit measurably below market effort because the market recruits for them.
Three practical consequences. First, retention investments compound beyond their direct effect — every point of attrition avoided also cheapens future hiring, a term absent from most retention P&Ls including, conservatively, the one above. Second, exits deserve grace: the departing engineer is tomorrow’s reviewer, referrer or boomerang candidate, and alumni programs repay their trivial cost. Third, the loop makes attrition a leading indicator for recruiting economics — a rising quarter shows up in next quarter’s funnel metrics before it shows up in finance’s headcount report. Centres that watch both dashboards together steer earlier, and steering early is most of what this article recommends.
Methodology & data notes
The trend chart traces the publicly disclosed attrition of India’s largest listed IT-services firms (gold) and the industry-reported captive band (teal); values are indicative annualised mid-points across quarters, harmonised for presentation — consult the IR archives for exact quarterly figures. The exit-driver shares are an indicative aggregate of published exit-survey reporting rather than a single study; the ranking, which is the claim, is consistent across sources.
References & further reading
- TCS Investor Relations and Infosys Investors — quarterly attrition disclosures, FY20–FY25
- NASSCOM — industry workforce and attrition commentary
- Gallup — State of the Global Workplace — manager effect and engagement research
- EY India GCC Pulse — captive-centre talent trends
Retention design is core to HexGn’s talent-activation practice — the diagnosis instrumented, the interventions sequenced by evidence, and built into the centre from week one rather than bolted on after the first exit wave.