EY says its 'invisible' AI router has helped cut token consumption by up to 60%

Jul 30, 2026 03:35 PM - 2 hours ago 4

EY does not fto each worker query spell consecutive to the smartest AI exemplary successful the room.

The Big Four firm has introduced an "invisible" router to thief negociate its soul AI spending, Dan Diasio, EY's world consulting AI leader, told Business Insider.

Rolled retired globally in April, the router sits down immoderate of EY's specialized AI devices and acts arsenic an intermediary to nonstop labor towards the champion AI exemplary for the task astatine hand.

"To the users, it makes nary evident change," said Diasio. They unfastened the frontier model, participate their prompt, and get a response. But down the scenes, the router is deciding wherever their query should go.

"Many times, it's faster than the frontier models, because to usage the heavier models intends there's much reasoning that gets put successful place, and it takes longer to beryllium capable to get backmost to the answer," Diasio added.

Token costs person go a increasing interest for companies arsenic AI providers progressively complaint based connected usage. From February to June, OpenAI, Anthropic, and GitHub introduced pricing models tied to tokens — the units that measurement AI's input and output — alternatively than pinch flat-rate billing.

The displacement has meant that using powerful models for elemental tasks tin quickly go costly.

EY US's AI Pulse survey, released Tuesday, recovered that 82% of elder leaders astatine companies investing successful AI were concerned astir token usage. The study was based connected responses from 534 elder decision-makers astatine US businesses betwixt April and May.

"A batch of these large bills that companies are getting amazed by are by the apical 1 aliases the apical 2% of group wrong the worker guidelines that are conscionable utilizing the incorrect instrumentality for the job," said Diasio.

A router helps prime the correct exemplary for the correct job, limiting token discarded and ensuring that much costly AI requests are reserved for the functions wherever they tin present the top value. It's a solution galore EY clients are besides building to thief negociate costs, said Diasio.

The router has not been deployed universally crossed each of EY's AI tools. Instead, it's been deployed connected department-specific platforms, including taxation and risk. It is not utilized for the wide Microsoft Copilot chatbot disposable to each EY staff.

The solution has had a "meaningful impact" connected really unit usage AI tokens, Diasio said.

The router, together pinch training and governance strategies, has helped trim token depletion by 60% since its implementation successful April, EY said.

The adjacent shape of AI monitoring

The first shape of the firm AI roar was astir getting labor to usage the exertion — cue the "tokenmaxxing" arena that encouraged workers to usage arsenic overmuch AI arsenic possible.

Rising costs person since shifted the attraction to controlling that use.

Corporate giants for illustration Disney and JPMorgan person installed dashboards to way employees' usage of AI, and AI-routing startups are connected the rise, helping developers nonstop tasks to different AI models, show for overspending, and quickly resoluteness outages.

EY has group token budgets based connected employees' roles and departments, Diasio said. Those who transcend their token allowance tin petition much done an support process.

Some 64% of elder leaders surveyed for EY's AI Pulse study said their organizations now show AI token usage and person clear budgetary guardrails for really overmuch they spend. Diasio expects that fig to emergence quickly: "If we inquire this mobility successful six months from now, it would astir apt beryllium for illustration 90% of people."

Dan Diasio headshot

Dan Diasio, EY's world consulting AI leader. EY

There are besides different levers that tin thief companies get much worth from their AI spending, Diasio said.

Businesses tin place the value of building beardown soul knowledge layers and designing amended prompts grounded successful their ain data. They should attraction their finance connected areas wherever AI tin person the deepest effect connected really activity is done, alternatively than spreading resources crossed each function, he said.

Monitoring worker usage will proceed to stay important, Diasio said, but corporate AI deployment is entering a caller shape successful which the attraction is connected worth and outcomes.

Fundamentally, measuring AI's worth requires looking beyond profit and loss, Diasio said.

Instead, he said, the broader questions businesses should beryllium asking include: Is AI enabling labor to get done tasks faster? And really tin you incentivize labor to research caller initiatives and responsibilities that aren't successful their existent occupation description?