EY developed an 'AI router' to manage its AI expenses and prevent them from escalating.

      EY has developed what they refer to as an “AI router,” a system designed to direct each task to the least expensive model capable of handling it, in an effort to manage their escalating AI expenses. According to a report by Business Insider, this tool represents the Big Four firm's solution to a growing issue within corporate IT: the costs associated with AI token consumption. The strategy is based on simple arbitrage principles.

      Not every request necessitates the most powerful or costly model; therefore, the router allocates simpler tasks to more economical models while reserving the more expensive ones for complex challenges, helping to reduce the costs without noticeably diminishing the output.

      EY has a significant reason to monitor these expenditures. The firm invests over $1 billion annually in AI, operates a fleet of approximately 1,000 AI agents, and has experienced a nearly 30% increase in AI-related consulting revenue—a scale where token costs become significant in a market where the most AI-focused companies spend thousands of dollars per employee each month.

      Their in-house research indicates widespread concern. In EY’s recent AI Pulse survey of 534 senior business leaders in the US, 82% expressed worry over token usage costs, and 98% of those using token-based tools stated that these expenses had prompted them to rethink their strategies.

      However, most companies lack proper oversight. Just 64% of the surveyed firms actively monitor token usage with budgetary constraints, meaning a third are incurring AI expenses without a clear tracking system—a situation that can lead to unexpected financial shocks within the industry.

      The sentiment has shifted from a focus on excess to sufficiency. Dan Diasio, EY’s global leader in AI consulting, stated clearly: “‘AI saves time’ is no longer adequate as costs increase and remain ambiguous,” illustrating the transition from adopting technology without regard to expense to pursuing value at a known cost.

      The economics of tokens are indeed perplexing. While the price per token has dropped as models have become less expensive, enterprise AI expenses have tripled due to the fact that agentic tools requiring multiple steps consume significantly more tokens than a single chatbot prompt.

      This is the challenge that the router aims to address. By aligning each task with the most cost-effective model that can still perform the required job, a company can continue utilizing AI extensively while preventing expenses from spiraling out of control—exactly what EY is attempting to demonstrate within its own operations.

      EY is not alone in this endeavor. The industry spent the past two years encouraging employees to maximize AI usage, a trend aptly named tokenmaxxing, but is now reversing course, with companies from Atlassian to Amazon implementing budgets and controls.

      For a consultancy, the router also functions as a product. EY provides AI consulting to other businesses, so a tool that effectively manages their own costs serves as a demonstration of their capability to deliver similar results for clients.

      The survey further underscores this trend. About 76% of leaders indicated that off-the-shelf software no longer meets their requirements, while 91% now consider in-house development of AI tools essential, favoring firms that offer the skills to create such tools.

      The challenge is that developing in-house solutions is complex. Nearly three-quarters of the leaders surveyed by EY reported that their own AI development initiatives were impeding progress, with a third identifying risks associated with shadow IT and governance issues—the messy reality behind the promise of a router.

      Ultimately, what this narrative signifies is a shift in perspective. The initial phase of the AI boom focused on whether a tool functioned properly; the second phase, to which EY’s router belongs, questions the cost of that tool and whether the resulting value justifies the expenditure.

      For now, EY’s approach is to create the metrics themselves. A firm investing a billion dollars annually in AI has determined that careful monitoring of each token is the path to maintaining spending, marking not a withdrawal from AI but rather an indication of how extensively it is now utilized.