Companies adopt smaller AI models as token costs squeeze ROI, says EY

Mahesh Makhija, Partner and Technology Consulting Leader, EY India

With token costs fluctuating rapidly and organisations struggling to quantify returns, companies are introducing token budgets, monitoring AI usage more closely, and increasingly turning to smaller language models (SLMs) for simpler tasks to keep costs under control while improving return on investment, according to an EY executive.“Every time a new model comes out, the price of the tokens is high. However, within 3-6 months, when a new competitor emerges, the token price collapses. It’s difficult to see the real return on investment from AI because of changing token costs,” Mahesh Makhija, Partner and Technology Consulting Leader, EY India, explained.Many companies are experiencing sticker shock as AI token bills climb, prompting closer scrutiny of why costs have risen, which models and agents are driving consumption, and whether those deployments are delivering sufficient value.Measuring AI value becomes a key priorityHe added that the most important thing is measuring the value of the tokens consumed. “For instance, in a certain part of EY consulting, suddenly, I see a $50,000 or $75,000 per month spend. This is fine if the consultants using it are seeing fantastic output. But can this be measured? Can I differentiate between people using AI properly and getting productivity versus those using AI and burning tokens?” he said.Makhija stressed that companies need to develop appropriate measurement frameworks to assess whether the tokens they consume are generating sufficient returns.Smaller language models emerge as a cost-effective alternativeTo combat this uncertainty, larger companies are resorting to token budgets and closely monitoring token prices. They are also moving towards smaller models, which will be more on-prem and easier to control.“You don’t need to use frontier models for everything,” he iterated. “The conversation naturally becomes that one doesn’t need a large model to do simple tasks. Enterprises are becoming smarter at using parts of open source. That is where the industry will trend also. In a way, SLMs can help combat token costs, and so, there may be a mushrooming of SLMs.”Published on July 10, 2026