The AI Value Office That Knows When to Leave

The Edge · Transformation · Monday, 27 July 2026

Researched by an agentic pipeline · reviewed and gated by the author

The best AI value office is therefore designed to become less necessary.

The portfolio that adds up to less than its parts

Each business unit can show progress. Sales has an AI assistant with rising usage. Finance has automated a reconciliation step. Operations has three forecasting pilots. HR can point to shorter drafting time. Every sponsor has a local benefit story.

Then the executive committee asks a simple question: what value did the enterprise receive?

The answers do not reconcile. Productivity was absorbed into more activity rather than lower cost or higher revenue. Several teams funded similar platforms. A workflow improved in one function only to create rework in the next. Shared model and data costs sit in technology, while claimed benefits sit elsewhere. No one has the authority to compare the whole portfolio and stop a popular local initiative in favour of a less visible cross-functional one.

This is the organisational problem behind the emerging AI value-realisation office. On 15 August 2026, Business Insider reported that EY is creating such an office to govern AI spending, monitor usage and returns, decide what scales and oversee workforce implications. [S1] The experiment is current and potentially consequential. It is not evidence that every enterprise needs another permanent corporate function, nor that EY’s office has produced results.

The useful idea is narrower: when value, cost and workflow ownership cross departmental boundaries, a temporary authority can integrate decisions that the existing organisation cannot yet make well.

The best AI value office is therefore designed to become less necessary.

The mismatch between value and authority

Most companies fund work vertically. Functions own budgets, people and delivery. AI value often travels horizontally. A customer-service agent draws on technology, data, legal, operations and workforce capability. An end-to-end claims redesign can reduce handling time in one team, change risk in another and shift human work into exception management elsewhere.

When authority remains vertical, three forms of leakage appear.

First, use cases multiply because every function can start but few people can stop. Second, shared foundations are underfunded because their value is distributed while their cost is concentrated. Third, local metrics reward activity—users, pilots, hours saved—without showing whether the enterprise converted that activity into growth, resilience, trust or lower cost-to-serve.

EY has advocated a value-realisation office since at least 2024, describing a unit that coordinates experimentation, resources, governance and benefits and may evolve into a C-suite control tower. [S2] Its more recent guidance argues for scaling through end-to-end processes, using fit-for-purpose outcomes, applying enterprise decision rights and managing AI as a strategic portfolio. [S3]

The logic is recognisable. It combines elements of a transformation office, portfolio board, centre of excellence and benefits function. The novelty lies less in the box on the organisation chart than in the economics it is being asked to integrate: variable model usage, shared platforms, agent autonomy, workflow redesign and workforce consequences.

The case against another office

The strongest objection should shape the design.

Finance already owns capital discipline. Technology already owns architecture and much of the cost. Transformation teams coordinate change. Business units own operating outcomes. If those functions are not working together, creating a new office can disguise the failure rather than resolve it.

The new centre may collect better data while lacking authority to change decisions. It may introduce an additional approval layer that slows local learning. Sponsors may learn to present benefits in the office’s preferred language without changing the work. Worst of all, business units may treat value as the centre’s responsibility: the office owns the dashboard, so the line no longer feels accountable for the outcome.

There is also a source problem. Current support for the model is dominated by consultancy research and self-reported surveys from organisations with an interest in AI transformation. PwC’s 2026 survey of 767 US operations and supply-chain leaders reports that 89 per cent gave at least one reason technology investments had not fully delivered expected results, while integration, data and adoption remained common barriers. [S4] That establishes a perceived value gap. It does not prove that centralisation closes it.

Any value office that cannot answer why existing governance is insufficient should not be created.

A temporary integration mandate

A credible office needs four tightly defined authorities and one explicit limitation.

Compare

Create a common economic and operational view across unlike initiatives. This does not mean forcing every use case into one ROI formula. Cost reduction, revenue, resilience, customer experience and trust have different evidence. It means using declared baselines, time horizons, confidence and benefit owners so that claims can be challenged on comparable terms.

Shared costs must be visible. Model usage, data preparation, security, integration, evaluation and human oversight are not free simply because they sit outside the sponsoring function’s budget.

Stop

An office that can recommend but not stop will become a reporting layer. It needs bounded authority to end duplicate pilots, pause initiatives with missing foundations and remove funding when evidence repeatedly fails. Stop decisions should be treated as learning, not embarrassment; otherwise sponsors will keep weak work alive to protect reputation.

Sequence

Local business cases routinely assume that platforms, data, skills and process ownership will arrive on time. The integrator must sequence shared foundations and cross-functional dependencies. A promising agent should not scale ahead of the identity, data or exception-management capability required to run it safely.

Scale

Scaling is not wider tool deployment. It is the redesign of an end-to-end value stream. Reuters reported EY leadership arguing that organisations do not gain ROI without changing jobs and processes, and that pilot proliferation can become a trap. [S5] The office should release scale funding only when a named process owner accepts the operating change and business outcomes.

Leave delivery ownership in the line

The office compares and arbitrates. It does not own the benefit. A sales leader still owns sales outcomes; an operations leader owns service and cost; a process owner owns the workflow. Central authority without line accountability produces cleaner governance and unchanged performance.

The value office should own the quality of the portfolio decision, never the fiction that it personally delivers every benefit.

What the mandate looks like in practice

Consider an illustrative group with 60 AI initiatives across five functions. The office begins with a nine-month charter.

It discovers eleven document-summarisation pilots using four vendors. Rather than declaring a winner immediately, it groups them by underlying job and data sensitivity. Six are stopped. Two remain local because their workflow is genuinely distinct. Three are combined behind a shared service.

The office then identifies a customer-onboarding redesign whose value spans sales, risk and operations. No single function can justify the integration cost from its own budget. The office sequences the common data work, allocates shared funding and requires one end-to-end process owner. It does not run onboarding. The business does.

At the third portfolio review, a forecasting tool reports high adoption but no improvement in inventory decisions. The sponsor argues that users like it. The office asks whether the forecast changed orders, stockouts or working capital against a baseline. Funding is paused pending a matched operational test.

By month nine, the useful question is not how many initiatives the office approved. It is whether the organisation can now make these comparisons without it.

The transfer test

Transformation offices often measure success through delivery volume: milestones, status, benefits claimed and issues closed. A temporary integrator needs a different final measure—capability transferred.

Four conditions indicate that central authority can shrink.

Transfer condition	Evidence
Comparable value discipline	Business cases use explicit baselines, confidence and benefit owners
End-to-end ownership	Named leaders control outcomes across functional hand-offs
Portfolio stopping works	Weak and duplicate initiatives end without exceptional escalation
Shared foundations are funded	Data, platforms, risk and skills have durable line budgets

Some shared services may remain central because scale makes that economical. The comparative authority does not need to remain a large permanent control tower. It can become a small portfolio capability within finance or transformation, while business units inherit the discipline.

BCG’s vision of an AI-powered transformation office offers a useful counter-model: automation can reduce coordination work and allow human teams to focus on intervention, while business leaders retain ownership. [S6] That design supports the central argument only if it strengthens the line rather than expanding the centre.

Evidence before institution

No public outcome evidence yet shows whether EY’s planned office changes resource allocation, work design or realised value. The announced mandate may evolve. Claims that most AI value is horizontal are interested-party findings, not universal laws. Organisational structure, business model and portfolio scale will determine applicability.

There are reasons for caution about value measurement itself. Reuters’ 2025 coverage of an EY survey reported that efficiency and productivity gains were often reinvested into more work rather than converted into immediate revenue or cost savings. [S7] A central office can make that distinction visible; it cannot decide that one form of value is inherently superior. Nor should every experimental initiative face a mature-business hurdle before it can learn.

The office should therefore be launched as a testable hypothesis. Give it a bounded portfolio, published decision rights and a fixed review date. Track duplicate work stopped, shared capabilities funded, decisions changed, processes redesigned and benefits realised. Compare decision speed before and after. Ask business owners whether accountability became clearer or migrated upwards.

If the office improves reporting but cannot show different decisions, it is an overhead. If it makes different decisions but the line does not deliver, it is a displaced accountability problem. If it changes allocation, enables cross-functional redesign and then transfers the discipline into normal governance, it has done its job.

The enduring capability is not the control tower. It is an enterprise that can see value across boundaries, stop what does not work and keep ownership where outcomes occur.

Sources

1. Business Insider — EY is creating a value realization office to make sure its AI spending pays off — 15 August 2026 — https://www.businessinsider.com/big-four-ey-creates-ai-value-realization-office-2026-8

1. EY — How can AI unlock value for industrials — 12 January 2024 — https://www.ey.com/en_gl/insights/advanced-manufacturing/how-can-ai-unlock-value-for-industrials

1. EY — How tech companies can break out of the AI ROI trap — 2026 — https://www.ey.com/en_us/insights/tech-sector/how-can-you-break-out-of-the-ai-roi-trap

1. PwC — 2026 Digital Trends in Operations Survey — 23 April 2026 — https://www.pwc.com/us/en/services/consulting/supply-chain-operations/library/digital-trends-operations-survey.html

1. Reuters — Firms must invest in real people to gain from AI, EY’s Teigland says — 22 January 2026 — https://www.reuters.com/business/davos/firms-must-invest-real-people-gain-ai-eys-teigland-says-2026-01-22/

1. BCG — The AI-Powered Transformation Office — 27 July 2026 — https://www.bcg.com/publications/2026/ai-powered-transformation-office

1. Reuters — Most companies suffer some risk-related financial loss deploying AI, EY survey shows — 8 October 2025 — https://www.reuters.com/business/most-companies-suffer-some-risk-related-financial-loss-deploying-ai-ey-survey-2025-10-08/