EY Built an AI Router That Picks the Cheapest Model for Each Task

EY, one of the world's four largest audit and consulting firms, has built a corporate AI router: a system that automatically routes tasks between language models and picks the cheapest one that is suitable. The tool was developed in response to fast-rising AI spending inside the company, Business Insider reported.

Why did EY need an AI router?

EY's AI router tackles one of the key corporate problems of 2025–2026: employees at large organizations default to flagship language models, even when the task doesn't call for one. The result is uncontrolled growth in token costs with a minimal gain in quality.

EY's system intercepts requests centrally and routes them on a "minimum sufficient model" principle: summarizing a document goes to a compact, inexpensive model, while legal analysis of a complex contract goes to a flagship. According to industry analysts, the cost gap between budget and top-tier LLMs ranges from 10 to 50 times per million tokens. That price gap is what makes model routing economically justified for organizations with a high volume of AI requests.

How widespread is this practice?

Model routing is not an EY invention: the approach has long existed in open-source tools such as LiteLLM and RouteLLM, and it is gradually being built into the enterprise gateways of major AI providers. What is new here is that a Big Four company built its own solution for its own needs rather than buying a ready-made product from a vendor.

• EY is one of the Big Four audit firms, alongside PwC, Deloitte and KPMG.
• The AI router was developed exclusively for internal corporate use.
• The system automatically selects the cheapest model capable of handling a given task.
• According to Business Insider, the tool is already in production inside the company.
• The problem of rising AI costs is recognized as systemic for corporate IT in 2026.

According to Business Insider, EY describes the tool as a response to "growing AI bills," a phenomenon that has become one of the main items on the CIO agenda.

"This is a problem that is now spreading across all of corporate IT," is the context

The Next Web cites for EY's development of its router.

What this means

EY illustrates a new corporate pattern: large professional services firms are no longer just adopting third-party vendors' AI tools, and are starting to build their own infrastructure for managing AI costs. Model routing is turning from a niche engineering practice into a basic element of a mature enterprise AI architecture. For most organizations, a concrete question follows: buy a ready-made routing solution from a vendor, or build your own, as EY did.