Faster Employees Are Not Making Companies Faster
The AI race is moving from usage dashboards to business audits, and India’s IT industry is caught on both sides of the shift.
Topics
News
- Meta Defends Child Safety Measures After Instagram Ad Row
- Google Cloud Plans More AI Data Centers in India
- Temasek To Lead $100 Million Funding Round in Pixxel: Report
- India May Regulate AI By Risk Under Proposed New Law
- UP Clears Data Center Policy, Targets $23 Billion Investment
- India Joins Inaugural UN Forum on Global AI Governance in Geneva
Image Credit- Chetan Jha/ MIT Sloan Management Review India
Key Takeaways
01
For two years, enterprises judged AI by tokens, prompts and active users. The bills have now outrun the measurable returns.
02
Individual speed does not equal company speed. Gains stall wherever AI sits outside the workflows and bottlenecks that govern real output.
03
By FY27 planning, boards should tie AI spend to decision quality and cost-to-serve. Indian IT firms must reprice around outcomes, not hours.
Some time this spring, engineers at Amazon Inc. worked out how to game an informal internal AI leaderboard. The board, called KiroRank, handed out points for AI usage. So employees began routing trivial tasks through expensive AI agents to run up their scores. One reportedly asked an advanced model to check the weather. The behavior got common enough that a senior executive, Dave Treadwell, reportedly told staff not to use AI just for the sake of using it. Soon after, the leaderboard quietly came down.
The mistake was not Amazon’s alone. For two years, a single number stood in for progress wherever generative AI reached software teams. More prompts meant more productivity. More tokens meant more output. More people logged in meant a company was pulling ahead. Engineers dubbed the habit “tokenmaxxing” and treated heavy consumption as proof they were innovating.
For Indian boardrooms, the reckoning cuts deeper than a runaway cloud bill. India does not only buy AI. It sells the human effort that AI is beginning to automate. An industry with more than $200 billion in annual IT exports grew around models in which revenue was closely linked to the number of people deployed and the hours billed. The same logic that is pushing enterprises to look beyond token counts is also pushing Tata Consultancy Services Ltd (TCS), Infosys Ltd, and their rivals to move beyond hours worked. India is being repriced from both sides of the invoice.
The Metric Was Only Ever an Input
“Token-maximizing isn’t the sign of a bubble. It is the symptom of an immature deployment strategy,” says Karan Kirpalani, Chief Product Officer at Neysa Networks, an AI cloud provider. Enterprises reached for the largest proprietary models because they performed best across most tasks. At that stage, counting prompts proved only that employees were experimenting.
“The industry did not optimize for the wrong metric but for the first available metric in an immature measurement environment,” says Biswajeet Mahapatra, Principal Analyst at research and advisory firm Forrester. Usage was an input. It was never an outcome.
Then the bills arrived. Microsoft Corp. began discontinuing most internal Claude Code licenses within about six months of rollout, shifting developers toward GitHub Copilot CLI instead.
Uber’s Chief Technology Officer, Praveen Neppalli Naga, told The Information the company had burned its entire 2026 AI coding budget in four months. Salesforce’s Marc Benioff has questioned whether every task belongs on the most expensive frontier model.
None of this happened because AI became more expensive per unit. Token prices fell by 60 to 80% across major providers over the past year. Consumption outran the discounts. Goldman Sachs expects agentic systems to multiply token demand roughly 24-fold by 2030. The unit cost fell, but the bill still rose.
The failure has an old name. When a measure becomes a target, it stops being a good measure. Told they would be judged by token counts, employees made the counts climb. The work did not follow. An MIT study last year found that 95% of generative-AI pilots produced no measurable profit-and-loss impact. Analysts have since contested how it defined failure. Read strictly or loosely, the gap between spending and return is real.
India comes to the same moment more cautiously. Nasscom’s latest adoption index finds that 67% of enterprises still spend under a tenth of their IT budget on AI. That restraint, long treated as a weakness, may now be a head start.
RESEARCH HIGHLIGHT This article draws on MIT Sloan Management Review India interviews with five executives across cloud infrastructure, process mining, workforce strategy and enterprise analytics. It is supported by disclosed FY26 results from TCS, Infosys, HCLTech and Wipro, and by secondary data from Nasscom, Forrester, Motilal Oswal, HFS Research and Axios. |
Individual Speed Is Not Organizational Speed
For Kaushik Mitra, Vice President and head of India go-to-market at process-mining software firm Celonis, the trouble lies in deployment, not the models. “The core RoI struggle stems from a severe operational context gap,” he says. General models reason well but rarely grasp a company’s processes, rules and dependencies. Employees paper over the gap with longer prompts and repeated requests, feeding compute without adding value.
Rahul Aggarwal, Vice President and General Manager at AI sales platform Proshort, puts the confusion plainly. Firms can count prompts submitted, summaries generated, and lines of code written. Those numbers show that people use AI. They say nothing about whether sales cycles shortened or decisions improved. “The right metric is not AI usage. It is the amount of business improvement AI helps generate for every dollar spent,” he says.
Chaitra Vedullapalli, Co-Founder and President of tech-inclusion nonprofit Women in Cloud, sees a deeper flaw in how enterprises imagine productivity compounding. Companies do not run as collections of independent workers. They run as systems, where one team’s output becomes another team’s input.
“At enterprise scale, productivity is systemic. One person’s output is another person’s input. Increase individual output without changing how work flows between people, and you create bottlenecks and misalignment.”
— Chaitra Vedullapalli, Women in Cloud
AI lifts individual capacity. Developers write code faster. Analysts finish reports in minutes rather than hours. But if downstream teams cannot absorb the extra work, or approvals stay slow, the enterprise moves at close to its old speed. The bottleneck has only relocated.
Factory managers learned this long ago. Speeding up one machine does nothing when the next station jams. Knowledge work is now meeting the same constraint. “Your job as a leader is to ensure that capacity is allocated toward problems that matter to the company, not just to the individual,” Vedullapalli says.
Not everyone reads the moment as a crisis. AWS Chief Executive Matt Garman says 90% of the CIOs he recently polled report positive returns or a clear path to them within months. The optimists and the skeptics may both be right. Returns show up where AI is wired into a workflow, and vanish where it floats beside one.
India Is Being Repriced From Both Sides of the Invoice
India’s technology companies sold effort. AI compresses effort. So the industry’s oldest annuity is deflating even as its AI revenue climbs. The pivot is already visible in how the largest firms talk about price.
At Tata’s annual meeting on June 9, Chairman N. Chandrasekaran called AI the biggest growth opportunity in TCS’s history and predicted the firm would soon run as many AI agents as it has employees. TCS’s annualized AI revenue crossed $2.3 billion by the end of FY26. Yet its own finance chief, Samir Seksaria, has warned of “revenue deflation” unless pricing moves from effort to outcomes.
Infosys shows the shift in its books. Fixed-price contracts accounted for 54% of revenue by March 2025, and Chief Executive Salil Parekh has openly said that AI’s gains invite outcome-based deals. Its engineers moved 3 million lines of Hertz’s COBOL code at 60% lower cost and on a 60% shorter timeline. A vendor advertising that discount teaches every future client what the work should now cost.
HCLTech is pushing in the same way. Chief Executive C. Vijayakumar advocates output-based pricing and expects 2 to 3% deflation in traditional services, even as advanced-AI revenue reached $620 million in FY26. Wipro has committed $1 billion to AI and is moving agentic work to outcome-based terms. LTIMindtree has gone furthest in form, launching BlueVerse Currency, a commercial framework built to price agentic work by result rather than headcount.
Clients are not waiting for renewals. HFS Research’s Phil Fersht says buyers are reopening contracts mid-term, often within 24 months, trading rate cards and offshore ratios for productivity commitments and gain-sharing. Motilal Oswal estimates that 12 to 15% of sector revenue faces direct AI displacement. The five largest Indian IT firms cut a combined 6,981 jobs in FY26.
As buyers, Indian enterprises have an opening that the West squandered. They are early, cautious, and backed by the ₹10,372 crore (about $1.1 billion) IndiaAI Mission, which subsidizes compute and lowers the cost of entry. Nasscom expects agentic AI to generate $300 billion to $400 billion in new service demand by 2030, across modernization, governance, and AI operations. That expansion rewards firms that price outcomes, not the ones that meter tokens.
India’s difference is not in how it uses AI. It is that AI is compressing the old services model even as it creates new demand. The country that learns fastest to sell outcomes will capture the growth. The one that clings to billed hours will absorb the deflation.
What Leaders Must Do Differently
C-Suite (CEOs, CIOs and Chief AI Officers) | Stop resourcing AI by seats and start wiring it into workflows. The executives interviewed agree that return depends less on picking the largest model and more on connecting AI to operational context, governance and business processes. Route routine tasks to cheaper models and reserve frontier models for hard reasoning. By the FY27 planning cycle, measure decision quality and cost-to-serve, not token counts. |
Functional Leaders | Redesign the process, not the task. A faster analyst helps no one if approvals downstream still take a week. Map where AI capacity actually lands, find the next bottleneck, and move the constraint before buying more compute. Track net business improvement for every dollar spent, and retire vanity metrics such as queries run. |
Boards & Governance | AI spending has become a capital-allocation question. The largest hyperscalers will spend roughly $650 billion on AI infrastructure this year. Boards should demand evidence that spending changes the P&L, not the dashboard. For Indian IT boards, the sharper duty is oversight of the pricing transition. A board that cannot say how fast its firm is moving from hours to outcomes cannot price its own deflation risk. |
The Scoreboard Comes Down
The leaderboards are losing their value, and the audits are starting. Usage still indicates whether a company’s people are experimenting. It does not show whether the business is improving. The companies that gain most from AI will not be those that use it the most, but those that change how decisions are made around it. India faces the harder version of the test. Its enterprises must learn to buy outcomes, and its IT industry must learn to sell them before clients set the price.
