AI Economics Maturity Model

Why a maturity model is needed

Many maturity conversations still focus on how much AI an organisation is using. That is not the same as whether the organisation can govern AI economically. An enterprise can have dozens of pilots and still be immature if it cannot explain where cost is gathering, who owns outcomes, or how value is being proved.

The research is consistent on this. MIT CISR finds that the most meaningful financial returns from AI appear when organisations move from scattered pilots into scaled, repeatable ways of working — not simply by deploying more technology. McKinsey's 2025 State of AI shows widespread AI adoption, but only around 5–6% of organisations reporting clear financial ROI. FinOps Foundation's 2026 data is precise: governance, forecasting, and organisational alignment rank as higher priorities than optimisation for AI scope — the signature of an industry still working on the foundational disciplines rather than refinement. IDC adds that only 7.5% of enterprises have meaningfully integrated FinOps into AI projects. These numbers describe an industry with broad experimentation and weak governance, not with weak technology.

The five levels

What each level looks like in practice

Level 1: Experimental

At Level 1, AI activity is energetic but economically thin. Costs sit in innovation budgets, software trials, local cloud accounts, or team-level experiments. Value is described in general terms such as productivity, innovation, or momentum.

Composite example: a diversified industrial group is running 27 AI pilots across sales, service, and engineering, but has no common inventory, no cost baseline, and no agreed owner for value realisation. The same workflow is being tested with three vendors in parallel.

Common pattern: the organisation confuses experimentation volume with strategic progress.

Priority actions for each role:

• FinOps Lead: create the first AI spend inventory across vendors, cloud accounts, and SaaS tools.
• TBM Lead: decide where AI spend will sit in cost pools and resource towers before more demand arrives.
• CFO: require a simple distinction between exploratory spend and operating spend.
• CIO: define whether a shared AI platform team exists or whether experimentation remains decentralised.
• Head of Engineering: document the largest technical demand drivers and duplicated experiments.
• VP of AI: establish a central register of pilots, sponsors, owners, and intended outcomes.

Level 2: Visible

At Level 2, reporting improves. Leaders can see major AI spend surfaces across tools, APIs, cloud capacity, and selected teams. The organisation can talk about cost, but not yet govern it with confidence.

Composite example: a mid-market financial services firm has consolidated its AI tool inventory and can now report about GBP 2.4 million in annual AI-related spending across 14 tools, three cloud platforms, and six API providers. But no single leader owns the AI cost envelope, and value is still described in terms of time saved per employee without baseline measurement.

Common pattern: the organisation mistakes improved reporting for maturity.

Priority actions for each role:

• FinOps Lead: begin reporting cost per inference or cost per workflow for the top AI services.
• TBM Lead: separate shared platform cost from local business demand in the service model.
• CFO: insist that AI funding requests include total cost of ownership, not only vendor prices.
• CIO: define boundaries between enterprise AI platform investment and business-unit experimentation.
• Head of Engineering: instrument the most expensive workflows for token, latency, and fallback visibility.
• VP of AI: map initiatives by owner, maturity, and expected return dimension.

Level 3: Managed

At Level 3, AI demand begins to be governed operationally. Shared standards emerge for routing, model usage, controls, reporting cadence, and workflow design. This is often where FinOps & AI and TBM & AI become formal parts of the operating model.

Composite example: a global retailer has established AI as a formal FinOps scope, reports monthly on spend by platform and product area, and can show cost per assisted customer interaction. It can forecast run-rate spend with reasonable confidence, but still struggles to show which use cases are producing durable business return.

Common pattern: cost discipline improves faster than value discipline.

Priority actions for each role:

• FinOps Lead: establish AI as a formal FinOps scope with a dedicated reporting cadence.
• TBM Lead: map AI costs into Technology Resource Towers using TBM Taxonomy 5.0 structures.
• CFO: require TCO-based review and expected return dimensions for major initiatives.
• CIO: formalise shared AI platform versus local experimentation boundaries and service levels.
• Head of Engineering: implement cost per inference and cost per action tracking at service level.
• VP of AI: maintain a central register of all AI initiatives with cost, owner, and status.

Level 4: Accountable

At Level 4, the organisation can explain what AI costs, who owns the target outcome, and what proof threshold applies before more scale is approved. The debate shifts from enthusiasm to evidence.

Composite example: a healthcare services organisation reviews AI initiatives quarterly by owner, baseline, cost envelope, risk class, and expected return dimension. Underperforming use cases are redesigned or stopped, while proven workflow automation receives additional funding.

Common pattern: initiative-level accountability is strong, but portfolio comparison remains uneven.

Priority actions for each role:

• FinOps Lead: connect unit economics to outcome reporting and scenario forecasting.
• TBM Lead: expose which capabilities and consumers are absorbing shared AI platform cost.
• CFO: introduce staged investment gates tied to proof and dependency level.
• CIO: govern platform investment through portfolio comparisons rather than one-off exceptions.
• Head of Engineering: tie performance and reliability metrics to economic thresholds for scale.
• VP of AI: standardise proof packs for executive review across all major initiatives.

Level 5: Optimised

At Level 5, AI is governed as a comparative portfolio. Leadership can compare cost structures, demand patterns, risk burden, and outcome evidence across multiple capabilities and redirect capital with confidence.

Composite example: a multinational services firm uses a portfolio review model to rank AI initiatives by strategic fit, cost-to-serve, realised value, and reuse potential. Shared platform investment is justified through portfolio economics, not through narrative alone.

Common pattern: success breeds complacency unless the organisation keeps adapting its model.

Priority actions for each role:

• FinOps Lead: continuously optimise AI unit economics and forecasting based on adoption curves.
• TBM Lead: use portfolio-grade cost models to support showback, chargeback, and scenario planning.
• CFO: compare AI capital allocation against other transformation bets using common standards.
• CIO: steer platform evolution based on reuse, resilience, and portfolio economics.
• Head of Engineering: embed cost-awareness into engineering design and release governance.
• VP of AI: rebalance the portfolio between predictable ROI, strategic innovation, and breakthrough exploration.

How to diagnose your current level

Use the same estate-wide questions across roles. Can the organisation distinguish shared AI platform spend from local demand? Can it explain cost per service or workflow? Are major initiatives assigned to named value owners? Is there a common proof threshold before scale? Can leaders compare initiatives against one another rather than only against their own business cases?

If reporting exists but ownership and proof remain weak, the organisation is typically between Levels 2 and 3. If ownership exists at initiative level but portfolio decisions still rely on narrative, the organisation is often between Levels 3 and 4. If portfolio comparison, stage gating, and capital reallocation are routine, Level 5 is in view.

What moves organisations upward

Five forces drive the transition upward: better visibility, clearer ownership, stronger operating discipline, higher proof standards, and more explicit portfolio governance. These are why The AI Value Gap, AI TCO Framework, and AI ROI Models connect so closely to this framework.

Capability progression matrix

Where most organisations sit today

The external evidence places most organisations between Levels 2 and 3. FinOps Foundation's 2026 data confirms that AI spend is now part of 98% of FinOps practices — a reasonable proxy for improved cost visibility. But governance maturity and proof discipline tell a different story. McKinsey finds only about one-third of organisations have scaled AI across the enterprise. IDC reports only 7.5% have meaningfully integrated FinOps into AI projects. IBM's study found just 29% of executives confident in measuring AI ROI. These figures describe an industry with strong Level 1 to Level 2 progress but limited Level 3 and beyond capability. Cost is being tracked. Ownership, proof standards, and portfolio governance are not yet functioning at comparable depth in most organisations.

Level transitions

Level 1 to Level 2

Trigger: AI activity becomes too widespread to treat as experimentation alone.

Hardest part: building a complete inventory across tools, platforms, and hidden budgets.

What takes longest: agreeing a common cost language across finance, technology, and business teams.

Level 2 to Level 3

Trigger: leaders can see spend and now need to influence the conditions creating it.

Hardest part: moving from dashboards to operating cadence, ownership, and policy.

What takes longest: instrumenting workflows well enough to produce useful unit economics and forecasts.

Level 3 to Level 4

Trigger: AI is becoming operationally important enough that cost control alone is insufficient.

Hardest part: assigning named value owners and enforcing proof thresholds before scale.

What takes longest: linking benefit claims to credible baselines and realised business outcomes.

Level 4 to Level 5

Trigger: multiple AI capabilities now compete for capital, platform capacity, and executive attention.

Hardest part: comparing initiatives systematically instead of letting each defend itself on its own narrative.

What takes longest: building portfolio-grade data and governance routines that can support capital reallocation with confidence.

Is Level 5 always the goal?

Most maturity models carry an implicit assumption that higher levels are universally better and that organisations should aspire to reach the highest level as quickly as resources allow. This assumption deserves challenge.

Level 5 governance — with a fully active portfolio comparison function, continuous optimisation, and portfolio-grade data infrastructure — is appropriate for organisations with a large, diverse AI portfolio where comparative capital allocation decisions have real consequence. For those organisations, the investment in Level 5 capability is justified because the decisions it enables are worth more than the governance overhead it creates.

For many other organisations, Level 3 or Level 4 is the economically optimal operating state. A £3M AI portfolio focused on a small number of proven use cases does not need the portfolio comparison infrastructure of an organisation managing £30M across forty initiatives. Over-investing in governance sophistication relative to portfolio size and complexity creates overhead without commensurate benefit — which is a form of misallocation as real as under-investing.

The framework is therefore better used as a diagnostic and a progression guide than as an aspiration in itself. The useful questions are: what level of governance is adequate for the scale and risk of the portfolio we are running? What specific capability is missing that is causing real management decisions to be worse than they should be? And what is the proportionate investment to close that specific gap?

An organisation with strong visibility and accountability but immature portfolio comparison capability should invest in Level 4 to Level 5 transitions. An organisation with immature visibility should invest in Level 1 to Level 2 transitions first — the more sophisticated governance layers are not useful without the foundational ones. And an organisation that has good Level 3 governance over a small, stable AI portfolio may be making the right call by not investing in Level 4 and 5 infrastructure until its portfolio size justifies it.

The framework is calibrated to help organisations become better governed, not to produce governance for its own sake.

Why this framework matters now

When AI is optional, weak economics can be tolerated for a while. When it becomes embedded in core workflows, customer journeys, software delivery, and planning processes, weak economics become a strategic liability. The Five Levels framework matters because it gives leadership teams a shared way to judge whether governance is improving fast enough to support scale — and whether the governance investment is proportionate to the scale being supported.

Conclusion

The most useful dividing line in enterprise AI is not between organisations that use AI and those that do not. It is between organisations that can govern AI economically and those that cannot. The first group can answer four questions with evidence: what does it cost in full, who owns the outcome, what does proof look like, and which initiatives should scale or stop? The second group answers these with narrative. The framework exists to make that gap visible — and to identify, specifically, which management capability is missing rather than leaving the answer as "we need better AI governance".