loading
loading
"Just because you consume tokens doesn't automatically mean there is a bank and you can start having a teller to withdraw money." That is how Satish Viswanathan sums up the thinking that took hold in 2024, when companies began treating token consumption as "a medal" of their AI capabilities. Some organisations even specified token consumption commitments in hiring packages. Two years later, companies are beginning to recognise that consuming tokens without any checks costs money without necessarily generating revenue. Viswanathan, the former global data and AI lead for travel at Accenture, is the author of The Weight of Intelligence. In an interview on Front Page with AIM Network, he argues that businesses are focusing too much on improving AI models and too little on their ability to absorb and use the technology. He calls this challenge "enterprise metabolisation". Having access to increasingly capable AI models does not automatically make an organisation more productive or profitable. Companies also need the people, processes, and infrastructure to put that intelligence to work. The Enterprise Adoption Problem Viswanathan identifies several factors that determine how effectively an enterprise can use AI: model capability, enterprise data architecture and availability, system integration and infrastructure, human processes and organisational structures, value economics, and governance and regulation. Every enterprise, he argues, has its own "technology debt, people debt, process debt" and organisational-structure debt. These constraints determine how much of a model's intelligence a company can actually use. Even access to artificial general intelligence (AGI), he believes, would not make every enterprise equally capable. "Still, every enterprise is unique." Companies cannot assume that the same technology will produce the same results everywhere. For business leaders, this means looking beyond the capabilities of AI models and examining whether their organisations can put those capabilities to productive use. Token consumption may show that employees are using AI, but it does not establish whether the technology is generating revenue or reducing costs. Measuring Financial Returns Viswanathan sees a shift in how companies approach AI projects. Competitive pressure and vendor partnerships are becoming strong tailwinds. He has also observed a pause in enterprise AI adoption. Companies are not necessarily abandoning the technology, but some are pursuing fewer use cases at once and reconsidering the scale of their investments. "We've already seen very, very smart pausing," he says, describing how some enterprises are containing their spending rather than going "full throttle". He linked this reassessment to pressure on chief executives to demonstrate returns from AI spending. As he put it, companies still need to figure out "what is the use case, what is the area of intervention and how to make money out of it". Productivity Gains are Not Savings Even when AI makes a task faster, the time saved does not automatically become a financial gain for the company. Viswanathan illustrates this with an employee who saves an hour by using AI to complete a task. The employee could, for instance, use that time for a yoga class. "We are not in this factory model where the moment we go at 9 am when you punch your card, you're invariably working," he notes. Knowledge workers do not necessarily spend every hour on assigned tasks, and time freed up by AI may not automatically be redirected to other work. The financial benefit depends on what the organisation does with that time. Companies need to encourage employees to use it productively, potentially by giving them new responsibilities or helping them develop additional skills. The same challenge applies to technology services companies. If AI allows a provider to complete work faster, clients may expect the resulting efficiency gains to be reflected in contract prices. The provider may therefore be unable to retain all the savings. Risks of Automation The pressure to improve efficiency also raises questions about how companies develop their workforce. Viswanathan warned that reducing entry-level hiring because AI can perform routine tasks could create problems over time. Employees build institutional knowledge through years of working on projects, observing decisions, and learning from failures. That experience helps prepare them for more senior responsibilities. "If you restrict people joining at the bottom of the pyramid, it's a disaster in the long term," he remarks. He acknowledges that reducing entry-level hiring could improve operating margins in the short term. However, companies risk weakening the pipeline through which employees acquire the experience needed to advance. As AI becomes part of everyday work, organisations must also decide how people and AI systems should share responsibilities. Referring to a Harvard study from roughly two years ago, Viswanathan describes two models of human-AI collaboration: the "cyborg" and the "centaur". In the cyborg model, human and AI contributions are intertwined, making it difficult to distinguish where one ends and the other begins. In the centaur model, responsibilities are divided more clearly between people and algorithms. Limits of Autonomous AI The question of human judgement becomes more important as companies adopt AI agents that can perform tasks with limited intervention. Viswanathan questions how readily businesses use the term "autonomous" to describe AI systems. Even experienced employees, he observes, seek a second opinion before making major decisions. Employees often consult colleagues or managers before acting. Having a human in the loop does not, by itself, solve the problem. Viswanathan describes human oversight as "your last resort". A person cannot realistically monitor thousands of agents or millions of interactions individually. Instead, he argues, enterprises will need observation models that monitor AI interactions, identify anomalies, and escalate relevant cases to people. "Otherwise, just saying ‘human in the loop’ means then you have to get more humans," he comments. That, he argues, could defeat the return-on-investment case for deploying AI in the first place. Redesigning Enterprises for AI Viswanathan's proposed response is to rethink how organisations make decisions and carry out work. He calls this approach "enterprise cognition", in which humans and AI jointly make decisions, with people retaining leadership and AI providing support. He does not advocate replacing an entire operating model at once. Instead, he favours an incremental transition wherein companies redesign one process or business unit at a time before extending the changes across the organisation. He cautions that simply adding AI to existing processes would not necessarily produce the financial impact companies expect. "If you're just doing ad hoc use cases... ad hoc processes being picked up and just AI-fied, you're not going to have that net P&L impact which we are hoping to have," he explains. The challenge for the leadership is understanding how AI can be incorporated into organisational decision-making and how those changes affect business performance. India's Opportunity Viswanathan believes India could strengthen its position in enterprise AI by drawing on its experience in technology services and systems integration. Achieving this, however, will require changes in education and research. Training programmes often focus on building AI solutions, he says. Enterprises also need people who can understand how to integrate those solutions into business operations and connect them to commercial outcomes. He calls for more research focused on practical industry problems. He also proposes an open-source ecosystem in which organisations could share knowledge and learn from one another while retaining ownership of their intellectual property. For India's global capability centres (GCCs), Viswanathan says the focus should shift from headcount to their economic contribution to the parent organisation's profit and loss. He wants India to become a centre for enterprise AI return on investment, helping companies translate AI capabilities into business value. He also highlights the importance of preserving human judgement as AI becomes more deeply embedded in work. In some situations, he suggests, organisations may deliberately limit AI use to help employees retain their critical-thinking skills.
document fetched 2026-10-26 02:58