In an interview with CNBC-TV18, Sylvain Duranton, Managing Director & Senior Partner at BCG X and Global Leader at BCG X, noted that businesses have moved past the debate over whether to adopt AI and are now concentrating on selecting suitable models for specific tasks while managing token costs effectively.
“The peak is ahead of us,” Duranton stated, emphasizing that he sees no indications of companies reducing their AI investments despite heightened scrutiny over returns.
His remarks are supported by BCG’s recent survey of 1,800 C-suite executives conducted prior to the World Economic Forum in Davos. The survey revealed that organizations expect to double their resources allocated to AI transformation between 2025 and 2026, averaging AI expenditure at about 2% of revenues.
The findings showed that 94% of CEOs intend to persist with AI investments even if they do not achieve anticipated returns by 2026. Additionally, 27% indicated they would boost spending, believing they had not aimed high enough.
“I don’t see any signs of hesitation,” Duranton remarked, adding that many CEOs increasingly regard AI transformation as integral to their long-term legacy.
As AI adoption accelerates, companies are also becoming more acutely aware of the cost implications associated with using large language models.
“One major concern is the expense of tokens, which is becoming increasingly tangible for many businesses,” Duranton noted.
Instead of relying solely on the largest commercial AI models, enterprises are progressively exploring open-weight and smaller models for routine tasks, reserving frontier models for more complex workloads.
“Currently, we find ourselves in a situation where everyone is using a fighter jet to pick up their kids from school,” Duranton commented, arguing that many enterprise applications don’t necessitate the most potent and costly AI models.
This shift reflects a broader transformation in enterprise AI strategy as organizations strive for the right equilibrium among performance, cost, and security. Depending on their market environments, some companies continue to favor Western AI models, while others are testing Chinese and open-weight alternatives.
Duranton highlighted that discussions have broadened beyond model performance to encompass issues such as safety, traceability, and cybersecurity.
India leads AI adoption, but monetising it remains a challenge
Although enterprises are swiftly adopting AI, translating this adoption into quantifiable financial returns presents the industry’s most significant challenge, according to Nipun Kalra, Managing Director & Senior Partner at BCG X and Global Leader of AI-Financial Institutions Practice.
Kalra noted that India is at the forefront of global workplace AI adoption, with around 95% of frontline employees utilizing AI multiple times weekly for work-related tasks.
“The positive news is that people are growing comfortable with AI. That’s encouraging. However, converting that into P&L gains for an enterprise is where the challenge lies,” he remarked.
BCG’s research also indicates that only 14% of organizations worldwide assess AI initiatives against specific profit-and-loss metrics, complicating the evaluation of whether investments are generating business value.
The leadership gap also poses a challenge. While 72% of CEOs globally personally spearhead AI initiatives, this figure drops to 55% in India, suggesting that many AI programs are still primarily driven by technology teams instead of business leaders.
Kalra argued that organizations treating AI as a business transformation initiative, rather than merely a technological deployment, are more likely to realize meaningful returns.
From AI pilots to enterprise-wide transformation
Kalra stated that many organizations are diluting AI’s impact by initiating numerous pilot projects instead of fundamentally redesigning their business processes.
“The organizations most likely to succeed are those that center on a limited number of priorities and completely rewire an entire function, customer journey, or business process from end to end,” he commented.
He further emphasized that enterprises need to build their internal AI capabilities instead of relying on consultants indefinitely.
“You may require an external catalyst to initiate and develop that capability. But if you lack that internal strength, it’s very challenging to continuously pivot, iterate, and enhance your core operations every day, month, and year. That internal capability is essential.”
Duranton added that AI is already transforming software engineering itself, with developers increasingly creating systems that generate, test, and validate code rather than writing every line manually.
“This is not an incremental change,” he stated.
For enterprises, the forthcoming phase of AI adoption is likely to be characterized less by experimentation and access to the largest models and more by selecting the most cost-effective model for each workload, developing internal capabilities, and demonstrating measurable business outcomes.
