AI Value Management Is Not FinOps for AI

It is the discipline of changing the economics of the enterprise

There is a category error at the centre of many conversations about AI value.

Because AI consumes cloud infrastructure, models, software, data and tokens, organisations naturally reach for the management disciplines they already know. They extend FinOps to AI workloads, add AI costs to IT Financial Management, map AI services into Technology Business Management, or place AI initiatives inside Strategic Portfolio Management.

All of those steps are useful. None of them defines true AI Value Management.

An organisation can optimise the cost of an AI system that should never have been built.

It can reduce cost per token while generating more content that customers do not value.

It can prove that an assistant saves employees five hours a week without converting any of those hours into additional revenue, lower operating cost, improved service or faster delivery.

It can deliver an AI programme on time, within budget and against its technical specification while weakening pricing power, increasing operational exposure or automating a process that should have been removed entirely.

These are not failures of cloud cost management, technology financial management or portfolio control. They are failures of enterprise management.

True AI Value Management is therefore not primarily about managing the economics of AI. It is about managing how AI changes the economics of the business.

That distinction moves its centre of gravity towards the CEO, CFO and COO.

Technology remains essential. It is the substrate on which AI value is built. But the decisive questions are increasingly commercial, operational, financial and organisational:

• What can the company now sell that it could not sell before?
• Which customers can it serve profitably?
• Which decisions can be made faster or better?
• Where can it create structural operating leverage?
• Which earnings can it protect?
• Which risks can it reduce, and which new risks is it creating?
• Which products, channels, roles and business models may no longer make sense?
• What should it disrupt before a competitor does?

Those questions cannot be delegated to a technology optimisation function.

This is not a criticism of FinOps, TBM, ITFM or SPM

The argument should not be built on a false distinction between disciplines that manage cost and a new discipline that manages value.

FinOps, TBM, ITFM and SPM have all moved towards business value.

The FinOps Foundation now defines FinOps as an operational framework and cultural practice that maximises the business value of technology, creates financial accountability and enables collaboration between engineering, finance and business teams. Its scope has expanded beyond public cloud into SaaS, licensing, data centres and other technology categories.

The TBM Council describes TBM as a framework for aligning technology investments and operations with measurable business outcomes. ITFM supports financial planning, forecasting, cost transparency and investment decisions across technology. SPM connects strategy, investment, resources and execution, helping senior leaders decide which initiatives to fund, adapt or stop.

These disciplines are not obsolete. A mature AI Value Management capability will depend on them.

The difference is not whether they care about value. The difference is the management object around which each discipline is organised.

There is substantial overlap. That is healthy.

The case for AI Value Management is not that existing disciplines are incapable of contributing. It is that none of them, by design, owns the full chain from AI capability to changes in revenue, margin, operating structure, risk, market position and enterprise behaviour.

FinOps can measure AI consumption and unit economics.

ITFM can build the financial plan and establish total cost.

TBM can connect AI platforms and services to business capabilities.

SPM can compare proposed investments and allocate resources.

AI Value Management must connect all of this to a harder question:

AI is not simply another category of technology spend

Cloud changed how organisations consumed computing.

AI can change what organisations produce, how decisions are made, how customers interact with companies, and whether human labour is required to perform parts of the work at all.

That is a larger economic shift.

AI is increasingly able to interpret information, generate outputs, recommend actions, coordinate activities and act across systems. As agentic systems mature, AI may execute meaningful portions of a business process rather than simply assist a person completing a task.

It may qualify customers, design offers, negotiate terms, write and test software, identify supply-chain disruption, adapt pricing, investigate fraud, generate products and coordinate other agents.

This makes AI more than an application layer. It begins to resemble a scalable form of cognitive capacity, one with unusual economic properties:

• it can be replicated at low marginal cost;
• it can operate across multiple functions;
• its capability can improve without rebuilding every process from scratch;
• it can augment labour, substitute for tasks, and create entirely new forms of output;
• it can make decisions and initiate actions at machine speed;
• its quality, cost and risk can change rapidly as models and markets evolve.

IBM's Institute for Business Value captures the strategic direction starkly: by 2030, AI may not merely enhance the business model, it may become part of the business model itself.

If that thesis is even partly correct, managing AI principally as a technology category is inadequate.

The three layers of AI economics

A useful way to distinguish AI Value Management is to separate three layers of economics.

Layer 1: Technology economics

This includes model, token, inference and infrastructure cost; licences and platform fees; data and integration cost; utilisation and allocation; performance, latency and reliability; architectural choices; vendor and commitment management; technical risk and control.

FinOps, ITFM, engineering, architecture, procurement and technology governance are naturally strong here.

This layer matters. Poor technology economics can destroy an otherwise promising business case. But efficient inputs do not prove valuable outcomes.

Layer 2: Operational economics

This includes throughput; cycle time; cost to serve; quality and rework; decision speed; service capacity; employee productivity; process resilience; headcount avoidance; capacity released or redeployed.

The COO, functional leaders, process owners, finance and workforce leaders become central.

This is where many AI programmes currently concentrate. They measure time saved, tasks completed and productivity improvements. Experimental evidence reviewed by the OECD shows that generative AI can produce meaningful task-level productivity gains, often between 5% and more than 25% in activities such as customer support, software development and consulting.

But operational improvement still does not automatically become financial value.

Layer 3: Enterprise economics

This includes revenue growth; conversion and retention; pricing power; gross margin; earnings protection; capital efficiency; market access; product differentiation; strategic option value; competitive position; business-model resilience; exposure to disruption.

This is the layer that determines whether AI matters to the enterprise.

The technology function may generate important AI value in its own right. BCG found that, in its 2025 study of 1,250 companies, the technology function accounted for 13% of measured AI value, almost twice the previous year's share. That is significant, but it also implies that most AI value sits outside the technology function.

AI may be funded through technology. It is not economically contained by technology.

Usage is not the unit of value

Cloud consumption created a strong relationship between usage, cost and optimisation. More compute, storage or data transfer usually creates more cost, making usage a useful management signal.

AI breaks that relationship.

High usage can mean strong adoption, useful automation, unnecessary generation, poor prompting, repeated failed attempts, low-quality content at scale, an uncontrolled agent loop, or a process producing more activity but no additional outcome.

Low usage can mean failure, but it can also indicate an exceptionally valuable intervention used only at critical moments.

A single AI-supported decision might prevent a major fraud, retain a strategic customer, identify a pricing error or avoid a regulatory breach. Millions of generated words may have almost no economic value.

Cost per token, cost per inference and active users are legitimate operational metrics. They are not measures of enterprise value.

The more useful unit is the economic intervention:

Productivity is only potential value

One of the most important distinctions in AI economics is the difference between capacity created and value captured.

Suppose an AI assistant saves 1,000 employees five hours each week.

The organisation can report 5,000 hours of productivity. It can assign a theoretical hourly rate and announce a large annual benefit.

But nothing necessarily changed in the financial statements, the customer experience or the company's capacity to grow.

The saved time may be absorbed by meetings and administrative complexity; spread across many people in increments too small to redeploy; used to produce more low-value work; offset by checking, correcting and governing AI output; retained as employee convenience; or consumed by higher expectations and growing workloads.

BCG's 2026 research makes this point directly: productivity gains alone do not translate into lower costs or better performance, and capacity does not become value until leaders decide how it will be redirected. Its June 2026 workforce survey found that 42% of regular frontline AI users reported saving at least eight hours a week, while most organisations had not established how to convert that time into value.

The conclusion is not that time savings are unimportant.

It is that time saved is an input into a management decision.

The value appears only when the organisation uses the capacity to increase output without proportionate cost; serve more customers; improve quality or responsiveness; accelerate product and revenue cycles; avoid planned recruitment; reduce contractor or outsourcing spend; remove work and simplify structures; redeploy people to higher-value activity; or release cost from the operating model.

True AI Value Management must own this conversion.

Otherwise, the organisation records productivity in a presentation while preserving the same cost base, structure and behaviours.

The five enterprise mandates of AI Value Management

AI value should be managed through a small number of economic mandates rather than an endless catalogue of use cases.

1. Grow revenue

AI can improve customer acquisition, conversion, personalisation, retention, cross-sell and share of wallet, pricing and offer design, sales productivity, product development, market reach, and speed from insight to commercial action.

McKinsey's 2025 global survey found reported AI revenue benefits particularly in marketing and sales, strategy and corporate finance, and product and service development. It also found that organisations achieving the greatest impact were more likely to pursue growth and innovation alongside efficiency.

The central measures are not users or prompts. They are commercial outcomes: incremental revenue, margin-adjusted revenue, conversion uplift, retention, price realisation, sales-cycle compression, revenue from new AI-enabled products, and time from concept to monetisation.

2. Create structural operating leverage

Efficiency should mean more than completing the same work slightly faster.

AI should improve the relationship between output and the resources required to produce it.

The strongest measures include output per employee, cost per transaction, cost to serve, unit margin, service capacity, process cycle time, defect and rework rates, working capital, and ability to scale without proportionate headcount.

This frequently requires removing steps, controls, handoffs and layers rather than inserting AI into an unchanged workflow.

The operating model must change around the capability.

3. Secure existing earnings

Not all AI value appears as new revenue or cost reduction.

AI can protect earnings by identifying customer attrition earlier, reducing fraud and leakage, preventing operational outages, improving supply-chain resilience, increasing compliance consistency, protecting service quality, detecting cyber threats, reducing warranty, claims or remediation cost, and identifying margin erosion.

This value is often invisible because it represents an event that did not occur.

It needs counterfactual measurement: what loss was likely, with what probability, and how much did the intervention reduce that exposure?

4. Mitigate new exposure

AI also creates risk.

Poor outputs, uncontrolled actions, bias, security failures, privacy breaches, intellectual-property issues and weak accountability can destroy value faster than infrastructure inefficiency.

NIST's AI Risk Management Framework and its generative AI profile emphasise the need to identify and manage AI risks in line with organisational goals, legal requirements, risk tolerance and priorities.

Risk should not be treated only as a compliance gate at the end of development.

Trust, control and resilience are economic variables. They affect adoption, brand, customer retention, regulatory exposure, insurance, cost of remediation, permission to scale, and the durability of value.

5. Create strategic options

Some of the most important AI investments will not generate an immediate, directly attributable return.

They may create proprietary data and feedback loops, reusable agent and workflow capabilities, organisational learning, ecosystem access, new distribution channels, customer insight, a path into a future market, or protection against disintermediation.

Economic research on general-purpose technologies shows why short-term measurement can be misleading. Brynjolfsson, Rock and Syverson's Productivity J-Curve argues that technologies such as AI require complementary intangible investment, including redesigned processes, skills and organisational capital, before their productivity benefits are fully visible.

This does not justify undisciplined investment.

It means the investment case should distinguish between realised value, evidenced leading indicators, capability value, strategic option value, and unsupported optimism.

AI Value Management should make uncertainty explicit rather than pretending every initiative has a precise, immediate ROI.

The real AI portfolio is not a portfolio of use cases

Most organisations currently manage AI as a list of assistants, copilots, pilots and agent projects.

That is administratively convenient and economically weak.

A use-case portfolio fragments the business into small technology interventions. It encourages local optimisation and makes it easy to count deployments without changing the system.

The better portfolio unit is the enterprise value chain or economic outcome: acquire and retain customers; design and launch products; price and sell; fulfil and serve; manage supply and inventory; detect and manage risk; allocate capital; develop and deploy talent; operate technology; manage the enterprise.

Within each value chain, leaders should consider the combined effect of AI, data, process design, automation, policy, organisational structure, skills, incentives, and human judgement.

The question changes from "Which AI use cases should we fund?" to "Which part of our economic system can now be redesigned, and what combination of changes will create the strongest outcome?"

This is a more demanding question because it removes the assumption that an existing process should survive.

AI may automate a task. It may also make the task, the role, the control, the management layer or the product itself unnecessary.

From business case to value contract

Traditional business cases are often approved before delivery and revisited only when funding is challenged.

AI requires a live value contract.

For each material intervention, the contract should define:

1. The enterprise outcome — What revenue, margin, capacity, earnings or exposure will change?
2. The baseline and counterfactual — What would happen without the intervention?
3. The accountable owner — Which business executive owns the result, not merely the delivery?
4. The value mechanism — How does AI capability produce the outcome?
5. The capture decision — What will management do with the capacity, insight or capability created?
6. The full economic cost — Include technology, data, integration, change, governance, workforce transition, quality control and ongoing operation.
7. The evidence hierarchy — Separate technical performance, adoption, operational impact and realised financial outcome.
8. The risk boundary — What decisions can the system make, under what controls, and with what tolerance for failure?
9. The scale and stop conditions — What evidence will cause the organisation to expand, redesign, pause or terminate the intervention?
10. The review cadence — How will assumptions be updated as models, costs, customer behaviour and competitors change?

The contract should remain active after deployment because AI value is dynamic.

A model improves. A supplier changes price. Competitors copy the capability. Employees adapt. Customers alter their behaviour. Regulation changes. A differentiated feature becomes a commodity.

AI Value Management is therefore not a stage gate. It is continuous economic steering.

A stronger evidence model

AI programmes often collapse several different types of evidence into one headline number.

A more credible approach separates five levels:

Most AI initiatives currently have technical and behavioural evidence.

Some have operational evidence.

Far fewer have repeatable enterprise-level financial evidence. McKinsey's 2025 survey found widespread use-case benefits and innovation effects, but only 39% of respondents reported EBIT impact at enterprise level.

This gap is exactly why AI Value Management is needed.

It is not another dashboard. It is the discipline that prevents a technical success from being automatically labelled a business success.

Where AI Value Management should sit

AI Value Management should not become another technology centre of excellence that asks business units to submit use cases.

It needs a federated operating model with enterprise authority.

The CEO owns the enterprise AI thesis

The CEO should define how AI is expected to change competitive position, business model, sources of growth, operating model, appetite for disruption, pace of investment, and strategic risk.

This is already becoming reality. BCG's 2026 AI Radar found that 72% of surveyed CEOs described themselves as the main decision-maker on AI, and half believed their job stability depended on getting AI strategy right.

The CFO owns economic integrity

The CFO should own baselines and counterfactuals, value classification, benefit validation, capital allocation, treatment of uncertainty, value attribution, the distinction between forecast, evidenced and realised value, and reinvestment and cost-release decisions.

BCG's June 2026 work on technology value argues that CIOs and CFOs need shared measures across operations, business expansion and disruptive innovation, rather than forcing every technology investment into one narrow ROI method.

The CFO should not be the department that slows AI down.

The CFO should be the executive who makes AI value credible.

The COO owns operational capture

The COO should own end-to-end process redesign, workflow and decision-right changes, capacity capture, structural simplification, service and quality outcomes, operating-model transition, and performance at scale.

This is where theoretical productivity becomes operational leverage.

Business leaders own the outcome

Revenue, retention, product adoption, service performance and margin belong to accountable business executives.

A sales leader cannot delegate conversion uplift to the AI team.

A product leader cannot delegate customer value to the data science function.

A service executive cannot claim time saved while leaving cost to serve unchanged.

The person who owns the P&L or business outcome must own the value contract.

Technology leaders own the enabling system

The CIO, CTO, Chief Data Officer and Chief AI Officer remain indispensable.

They should own architecture, platforms, data readiness, model selection, engineering, technical performance, technology economics, security, integration, lifecycle controls, and technical scalability.

They enable enterprise value. They should not be expected to manufacture commercial outcomes that sit outside their authority.

The CHRO, risk and legal functions shape the economic result

AI value is inseparable from workforce design, incentives, skills, employee trust, accountability, regulatory permission, customer confidence, and risk tolerance.

The CHRO, Chief Risk Officer, General Counsel and security leadership are not support participants. They influence whether AI value can be captured and sustained.

The emerging operating model

A practical AI Value Management structure may include:

1. An enterprise AI value council

Sponsored by the CEO, with the CFO and COO as permanent leaders.

Its job is to allocate attention and capital across enterprise value pools, resolve cross-functional barriers and make decisions that individual technology or business teams cannot make alone.

2. A small AI value office

This should not own every initiative.

It should maintain the enterprise value taxonomy, value contracts, evidence standards, benefit and risk assumptions, portfolio comparisons, value-realisation reporting, and methods for counterfactual and option-value analysis.

3. Embedded value owners

Each major value chain or business area should have an accountable owner who can change process, people, policy and investment, not simply sponsor a technology deployment.

4. Enabling disciplines

FinOps, TBM, ITFM, SPM, enterprise architecture, data governance, responsible AI, security and change management should provide the specialised capabilities required to make the system work.

AI Value Management should orchestrate these disciplines, not absorb or replace them.

A proposed AI value equation

A complete AI value equation cannot be reduced to benefits minus model cost.

A more useful formulation is:

Realised AI value = incremental revenue + protected earnings + released or redeployed capacity + avoided loss and exposure + strategic option value - full lifecycle cost - transition and organisational friction - quality, trust and risk failures - cannibalisation and second-order consequences

Not every component can be measured precisely.

That is normal.

Boards and executive teams already make decisions involving forecasts, probabilities, strategic options, brand, market response and uncertain future cash flows. AI should be held to evidence, but not restricted to whatever can be extracted from a billing file.

The goal is not false precision.

It is disciplined judgement.

The uncomfortable implications

True AI Value Management will force decisions that many organisations would prefer to avoid.

It may show that high adoption has produced little value; productivity has created capacity that management has not captured; a successful assistant should lead to role or process redesign; an efficient AI service supports a weak product; a profitable current offering should be cannibalised; the company must invest before the return can be measured conventionally; some AI programmes should be stopped despite strong executive enthusiasm; some functions will need to lose control for an end-to-end value chain to improve; governance can create value, while excessive control can destroy it; or the biggest risk may be moving too slowly rather than spending too much.

This is why the discipline cannot be owned solely by technology.

The difficult choices concern customers, products, people, capital, accountability and the future shape of the enterprise.

The long-term conclusion

The greatest mistake would be to turn AI Value Management into the next technology optimisation function.

That would be comfortable.

It would produce dashboards, taxonomies, unit costs, maturity assessments and governance meetings.

It would also risk missing the point.

FinOps can show whether AI is being consumed and operated economically.

ITFM can show what it costs and how that cost will change.

TBM can connect AI resources and services to business capabilities and outcomes.

SPM can help select and govern investments.

Responsible AI and risk frameworks can define the conditions under which systems should operate.

All are essential.

But true AI Value Management begins with a different premise:

Its purpose is not to prove that AI has value.

Its purpose is to determine where AI changes the economic equation, decide what management must do to capture that change, and continuously move capital and attention towards the strongest evidence.

That places it close to the CEO, CFO and COO because it is ultimately concerned with growth, operating leverage, earnings, exposure and strategic position.

AI Value Management begins where technology value management reaches the edge of its authority.

In time, the term itself may disappear.

It will simply become how the enterprise is managed.