EY turns AI cost control into a standing office, and claims 60% fewer tokens for it

EY has created an AI Value Realization Office to centralize oversight of AI investment and connect that spending to measurable business impact, and Global Chief Innovation Officer Joe Depa says the combined program of model selection, employee training and usage governance cut the firm's overall token consumption by 60% while delivered value increased ^[1][2][3]. That pairing, a permanent office with authority over which models get used plus a large headline savings number, is the template other large organisations will be handed by their own consultants.

The remit is broader than procurement. According to Dan Diasio, EY's global consulting AI leader, the office will govern AI spending, monitor usage, decide which initiatives to scale, and assess effects on the workforce ^[5]. Bloomberg reported on August 10 that EY is hiring a head of "agent economics" to run it and give firm leadership a consolidated view of where AI investment produces tangible impact ^[6]. Business Insider reported on August 15 that the function should be fully operational within a couple of months, which puts the target somewhere around mid-October ^[4][14].

The stated reason for centralizing is that the money is already spread out. Diasio told Business Insider that AI investment cuts across IT, finance, sales, HR and operations, which makes department-by-department budgeting a poor fit for enterprise-wide opportunities ^[7]. He cited EY-Parthenon research assigning 75% of potential enterprise AI value to horizontal value streams spanning multiple functions, against 25% from projects contained inside a single function, a three-to-one split ^[8][15]. If that holds, the awkward implication for anyone copying this is that the departments funding AI today are the ones least positioned to capture the return.

Read as an org chart change rather than a technology one, this is a capital-allocation mechanism wearing a governance label. Diasio says the office will look beyond direct financial return to whether initiatives actually change business performance, and will steer funding toward the largest opportunities ^[9]. In practice a function like that ends up requiring teams to report model usage, inference cost, output quality, adoption, human-review rates and business outcomes on a common definition ^[12]. That consistency is the point, because a local productivity gain often does not survive once shared infrastructure cost, workflow rework and ongoing human oversight are charged against it ^[13].

Two things to keep in proportion. The 60% token reduction is EY's own figure, attributed to several interventions at once, not an independently audited result or a benchmark anyone can reproduce ^[11][3]. And token consumption is an input measure; cutting it is compatible with better routing to cheaper models, tighter prompts, or simply less usage, and the source material does not separate those. EY has not publicly specified the office's full measurement framework, its decision rights, or its technology stack ^[10].

What to watch: whether the office is actually running by the autumn target ^[4], who is hired into the agent economics role and what authority that seat carries ^[6], and above all how EY defines value when it has to compare unlike initiatives and shut one down. An office that can only approve and scale is a budget line. One that can stop a system already in production is governance. The first published example of the latter is the evidence worth waiting for.