Revenue equals headcount. That is the entire model.
TCS generates $49,902 per employee. Infosys: $60,164. Wipro: $45,118. Compare that to Apple at $2.4 million per employee, Google at $1.95 million, Microsoft at $1.08 million. The gap runs 20-50x. This is not a difference in degree. It is a difference in kind.
Technology companies scale through code. IT services companies scale through bodies. Employee benefit expenses consume 57-58% of revenue across the top firms. Every quarterly earnings call reports "utilization rates" - the percentage of employees actively billing clients, typically 83-86%. Software product companies do not report utilization rates. That metric only exists in businesses that sell person-hours.
Revenue per employee in Indian IT rose from roughly $30,000 in the 1990s to $50,000 today - approximately 1-2% annual growth over 25 years, barely keeping pace with inflation. That is what a labor business looks like at scale: 1.56 million people across the top five firms, generating roughly $80 billion in combined revenue, with margins determined by the spread between what they pay engineers in Bangalore and what they bill clients in Boston.
"Our assets walk out of the door each evening."
Infosys's founder articulated the deepest vulnerability with characteristic directness. Vineet Nayar at HCL wrote an entire Harvard Business Press book on this premise. TV Mohandas Pai, former Infosys board member, quantified the damage at $5,000 per departing employee in direct costs alone - recruitment, training, and six months of non-billable ramp-up time.
During peak attrition in FY2023, Infosys hit 28.4% annual attrition. Wipro reached 23.3%, HCL 23.8%, even TCS hit 19.7%. The top seven Indian IT companies made 380,000 replacement hirings in a single year - roughly 1,040 new hires per calendar day. At Pai's estimate, that represents $1.9 billion in annual replacement costs for the top firms alone.
And that figure captures only direct costs. It misses the client relationship nuances, the system quirks learned over years of maintenance, the debugging intuitions that no runbook captures. All of that vanished with each resignation letter.
Billions spent on knowledge management. Knowledge still walks out.
By 1999, 15 of the world's 23 CMM Level 5 companies were Indian. TCS built the world's largest corporate learning center on 97 acres - capacity for 15,000 people simultaneously, 50,000 annually. Infosys invested over 2,000 crore in its Mysore campus, training 14,000 people at a time across 147 training rooms.
Each company built proprietary knowledge management systems. Infosys created KShop with "Knowledge Currency Units" - virtual currency for contributing knowledge, redeemable at company gift shops. The incentive system led to gaming and quality degradation. Groups abandoned KShop entirely and built shadow repositories. When monetary rewards were adjusted, new contributions collapsed within six months.
The failure ran deeper than execution. Michael Polanyi established in 1966 that "we can know more than we can tell." Research consistently estimates that 80-90% of organizational knowledge is tacit - embedded in intuitions, relationships, and experiential understanding. KM systems operate almost exclusively on explicit-to-explicit knowledge transfer, which represents at most 10-20% of what an organization actually knows. Globally, $73 billion was spent on KM software with, as one researcher concluded, "few and far between" examples of success. The BT research director estimated KM failure rates at 70%.
1,700 GCCs are hiring your people. And paying them more.
India now hosts over 1,700 Global Capability Centers, employing 1.9 million professionals - comparable to the combined headcount of TCS, Infosys, Wipro, and HCL. These are not cost-center back offices. Over 50% have evolved into transformation hubs. Seventy-five percent of GCCs established since 2021 are R&D-driven.
GCCs pay 12-30% salary premiums over IT services companies, with AI/ML specialists commanding 30-50% premiums. Critically, 71% of GCCs now offer ESOPs, RSUs, or stock appreciation rights - long-term incentives that increase total compensation by 20-100% and are rarely offered in IT services.
As one GCC executive told Analytics India Magazine: "Our GCC is taking business away from TCS/HCL and replacing them or taking their staff and making them ours." IT services companies invest months training freshers at massive campuses, then watch those newly skilled employees get hired away by GCCs - sometimes by the very same clients the services firm was serving. GCC hiring has outpaced IT services hiring since early FY2022, resulting in 20-25% lateral movement from services to captive centers.
Only 3.2% of organizational knowledge survives a year.
Linda Argote's landmark research at Carnegie Mellon quantified what services leaders experience intuitively. Studying Liberty ship production during WWII across 13 shipyards, she found that knowledge depreciation is devastatingly rapid: only 3.2% of an organization's knowledge stock survived after one year.
Benkard's analysis of Lockheed L-1011 TriStar production confirmed a monthly depreciation rate of 0.96, meaning roughly 39% of accumulated experience vanished annually. David and Brachet, studying ambulance companies, determined that 62% of organizational forgetting is directly attributable to labor force turnover.
Arnold Kransdorff coined the term "corporate amnesia" in 1998 - organizations literally forgetting how to do things as people leave. He estimated the cost at up to 9.7% of GDP. For IT services companies running at 12-15% attrition even in stable years, the math is unforgiving. These organizations lose and rebuild a substantial fraction of their institutional knowledge base every single year - a Sisyphean cycle masked by process documentation that captures the what but never the how or the why.
AI captures what knowledge management never could.
Erik Brynjolfsson's "Generative AI at Work" study, published in the Quarterly Journal of Economics in May 2025, provides the most rigorous evidence yet. Studying 5,172 customer support agents at a Fortune 500 firm, Brynjolfsson and his team found that AI-assisted novice workers improved productivity by 34%. Workers with less than one month of tenure saw a 46% improvement.
The critical finding: AI-treated agents with just two months of experience performed as well as untreated agents with over six months. The experience curve compressed by two-thirds. The AI system learned patterns from top performers and made those patterns available to everyone - doing precisely what decades of KM systems failed to do. It captured not what experts documented, but how experts actually behaved.
Foundation Capital's 2025 analysis articulated the concept of "decision traces" - the reasoning that connects data to action, which "was never treated as data in the first place." When AI systems operate in the execution path, every interaction creates a persistent record of how context turned into action. These accumulate into a living, searchable record of institutional decision-making. The exception logic that lives only in people's heads. The precedents from past decisions. The cross-system synthesis that happens in a support lead's mind.
A separate study achieved 94.9% full-knowledge recall through LLM-based agent frameworks that iteratively reconstruct tacit knowledge through structured employee interactions - a result that would have seemed impossible under traditional KM paradigms.
The divergence is underway. The window is closing.
India's top five IT services firms lost over $150 billion in combined market capitalization in 2025. TCS alone shed nearly $70 billion. The Nifty IT Index fell 20% in a single month - its steepest decline since the 2008 financial crisis. Jefferies estimates AI could shrink the managed services business, which accounts for 22-45% of revenues across leading Indian IT companies.
Brynjolfsson's research on the productivity J-curve shows that early AI adopters experience initial dips but emerge with significantly stronger growth trajectories. Firms that delay enter their J-curve dip precisely when early movers are accelerating - a compounding disadvantage. The OECD documented this divergence empirically: frontier firms grew labor productivity at 2.8% annually versus 0.6% for laggards, with the gap even wider in business services.
For IT services CEOs, the calculus is stark. Every departing employee permanently subtracts institutional knowledge. Every day without AI-enabled knowledge capture is another day of irreversible loss. Competitors that capture institutional knowledge will deliver more consistent service, onboard replacements faster, and suffer less from the turnover that plagues the industry structurally.
Not a consulting firm. A transformation builder.
Gaurav Rastogi wrote the book on how this industry was built - literally. Offshore (Penguin, 2011) documented the rise of Indian IT services. Global Business in the Age of Destruction and Distraction (Oxford University Press, 2022) examined what happens when that model meets structural disruption.
The Retained Learning Thesis is the answer to both books: an architecture that turns the knowledge-loss problem from something you manage into something you eliminate. It is running in production right now - not a pilot, not a deck - with 3,200+ commits in the last year.
Visiting faculty at IIM Ahmedabad and Ashoka University. Board member at GTU and HSCI Global. Twenty-five years inside the industry this paper describes.
doloopdigital runs retained learning transformations. Architecture advisory, build, and deploy. We work with services businesses and their PE partners to prove the compounding curve in production.