Why AI Intangible Assets Defy Board Valuation

The board’s AI value problem

AI intangible assets are becoming a balance-sheet paradox for boards in 2026. Proprietary models, curated training data, workflow logic, evaluation libraries, and the operating capabilities surrounding them can influence revenue, margin, and resilience, yet associated spending may appear as current-period expense. That visibility gap can obscure the difference between durable enterprise value and experimentation that has not earned the right to scale. Directors need a disciplined view of both without treating every AI initiative as a capital asset.

At board level, the issue is not whether AI is important; it is whether the organization can explain which investments create defensible advantage, which improve execution, and which remain hypotheses. Traditional reporting can aggregate model development, vendor fees, data preparation, training, redesign, and change management into broad expense lines. A strong quarterly review should connect spend to operating outcomes, expected life, ownership, and explicit stop-or-scale decisions.

That discipline is particularly important when management presents an AI program as a platform rather than a collection of projects. A platform label can imply permanence, but it does not establish control, useful life, or recoverable value. Directors should ask whether the organization could operate, maintain, and defend the capability without the original vendor, team, or data source. If the answer is no, the dependency should be visible in the investment case and risk reporting.

AI intangible assets expose the limits of conventional reporting

AI intangible assets do not sit in one neat category. Treatment depends on applicable accounting standards, legal control, development stage, technical feasibility, and expected benefits; it cannot be determined by calling something a platform or asset. Some costs may qualify for recognition, while others must remain expense. The board’s role is not to rewrite policy. It is to require a consistent map from economic resource to accounting conclusion, valuation assumption, and disclosure.

• Models and code. A model may be proprietary, but its value depends on performance, integration, maintainability, and the cost of reproducing it.
• Training data and rights. Data quality, provenance, permitted use, exclusivity, freshness, and remediation obligations determine whether data supports durable advantage.
• AI-enabled workflows. The value may reside in redesigned decisions, controls, and integrations rather than in the model alone.
• Organizational capability. Talent, governance, evaluation routines, and learning loops can create repeatability, but individual labor remains different from a controlled resource.

This decomposition gives boards a bridge from cash spent to controlled resource to operating outcome, without pretending that every productive capability belongs on the balance sheet.

Separate the AI stack before assigning value

Valuation starts with separation. Management should define a unit of account, owner, dependency, expected benefit, and evidence for each layer before selecting a cost, market, or income approach. A cost view may inform reproducible software; an income view may better reflect a workflow with measurable cash flows. Neither method can compensate for weak evidence.

• Core intellectual property. Separate reusable model assets, evaluation methods, and code from routine configuration.
• Data assets. Record provenance, permissions, exclusivity, refresh requirements, and restrictions on commercial use.
• Workflow assets. Measure embedded controls, integrations, user adoption, and process redesign.
• Capability assets. Identify repeatable operating routines while distinguishing them from ordinary personnel expense.

An AI asset register should link each item to an accountable executive, investment history, business process, performance indicators, legal dependencies, and an expected review date. That register becomes the common source for finance, risk, technology, and strategy rather than allowing each function to describe value differently.

Separation also prevents double counting. The same benefit should not be attributed independently to a model, a workflow, and a transformation program if each relies on the same incremental cash flow. A clear dependency map lets finance test whether value is additive and lets directors see where a failure would propagate across the portfolio. This is especially important where third-party infrastructure or licensed data creates a concentration risk.

A five-part test for AI intangible assets

The TRACE framework gives directors a repeatable review of AI intangible assets. Its five phases force management to connect strategic claims with accounting discipline, evidence, and decision rights.

• Trace economic contribution. Map each resource to revenue growth, cost reduction, risk reduction, service quality, or strategic resilience. Require a baseline and a measurable counterfactual.
• Reconcile accounting treatment. Document the relevant policy, cost boundary, development stage, and recognition conclusion. Reconcile the management view with the general ledger.
• Assess durability. Test performance stability, adoption, data rights, switching costs, replaceability, and dependence on scarce people or external providers.
• Challenge the investment case. Review base, upside, and downside scenarios, including operating costs, adoption friction, regulatory constraints, and credible alternatives.
• Establish controls and disclosure. Assign owners, approval thresholds, monitoring duties, impairment triggers, and the disclosures required when judgments become material.

The output is not a single invented number. It is a defensible range, an evidence trail, and a clear statement of what must be true for value to persist.

Apply the TRACE framework to capital allocation

TRACE should be built into the investment cycle, not reserved for annual reporting. The board can require management to use the same evidence standard when approving a pilot, scaling a workflow, or renewing a major data and model commitment.

• At funding: require a concise thesis, owner, baseline, spend ceiling, and intended decision date.
• At pilot exit: require observed user behavior, quality results, control performance, and evidence of economic benefit.
• At scale: approve capacity, data, and change investment only when adoption and unit economics meet defined thresholds.
• At portfolio review: compare initiatives by strategic contribution, risk, reuse, and evidence rather than by spend already incurred.

This creates an options discipline: funding is staged by evidence, not sunk cost. It also produces a more credible portfolio view of AI intangible assets and makes it easier to redirect capital without labeling every stopped experiment a failure.

AI intangible assets require a different impairment lens

AI intangible assets can lose value without becoming visibly obsolete. A competitor can change user expectations, a data license can narrow, model quality can decay, or run costs can erase margin. Initial valuation must therefore be separated from ongoing recoverability. Under applicable standards, management should identify testing units and triggers; directors should challenge whether forecasts reflect observed behavior rather than aspiration.

• Commercial evidence: monitor adoption, retention, pricing power, contribution margin, and substitution by simpler tools.
• Technical evidence: track drift, accuracy, latency, security incidents, control failures, and the cost of retraining or replacement.
• Rights and dependencies: review licenses, data provenance, vendor concentration, talent concentration, and changes in permitted use.
• Strategic evidence: test whether the capability still supports the stated advantage when competitors, regulation, or customer behavior changes.

The audit committee should require a trigger dashboard and a documented response protocol. A missed forecast is not automatically impairment, but repeated misses without a credible recovery path should not be concealed by optimistic terminal assumptions.

Stress-test the portfolio before it scales

Board approval should be based on scenarios rather than a single forecast. For each material initiative, management should show the value if adoption is slower, input costs rise, performance falls, or a key permission is withdrawn. The analysis should identify which benefits are contractual, which depend on customer behavior, and which can be replaced by a lower-cost alternative. It should also show the cash required to maintain the capability after launch, including monitoring, retraining, security, governance, and change management. A business case that ignores those recurring obligations overstates economic value.

• Adoption downside. Test delayed rollout, uneven usage, and additional training or process redesign.
• Cost downside. Model inference, storage, data-refresh, specialist-talent, and control costs at higher volumes.
• Replacement case. Compare the proposed capability with a vendor service, conventional automation, or a simpler process.
• Resilience case. Assess whether the initiative still matters if a supplier, data source, or model becomes unavailable.

Scenario analysis gives directors a better basis for sequencing investment. It also clarifies whether management is buying a durable capability or renting performance from a changing external ecosystem. That distinction should influence contract terms, exit rights, contingency plans, and the size of any committed spend.

Turn disclosure into a decision-useful signal

Disclosure should help investors and directors understand where AI creates value, where judgment is significant, and where uncertainty is concentrated. When AI intangible assets are material to strategy or performance, generic statements about innovation are insufficient and can reduce confidence rather than increase it.

• Materiality: explain the assets, commitments, and exposures that could affect enterprise value.
• Judgments: describe useful lives, recognition boundaries, valuation methods, and key assumptions.
• Economics: connect investment to adoption, margin, productivity, resilience, or other relevant outcomes.
• Controls: summarize ownership, model risk oversight, data rights, and significant weaknesses or dependencies.

Useful disclosure does not require revealing sensitive model details. It requires enough specificity for a reasonable reader to distinguish repeatable capability from unproven experimentation and to understand how management would respond if assumptions fail.

Directors should distinguish a metric that demonstrates activity from one that demonstrates value. The number of models launched, prompts processed, or employees trained may show momentum, but adoption quality, avoided loss, realized margin, and retention are more informative for capital allocation. Reporting should also identify unresolved assumptions rather than presenting precision that the evidence cannot support.

Build governance around evidence, not enthusiasm

Governance fails when responsibility is fragmented across technology, finance, legal, risk, and business units. The board should establish one accountable executive for the portfolio, while preserving independent challenge from finance, internal audit, risk, and the relevant board committee.

• Set decision rights: define what management can approve and what requires committee or full-board review.
• Use common metrics: standardize adoption, quality, unit cost, benefit realization, incidents, and dependency measures.
• Protect challenge: require independent validation of material models, data rights, controls, and valuation assumptions.
• Review people risk: monitor key-person dependency, capability gaps, succession, and incentives tied to optimistic outcomes.

The board should also ask whether compensation rewards durable benefits or merely the launch of AI initiatives. Incentives should favor realized value, responsible scaling, and timely decommissioning.

What directors should decide in 2026

By the 2026 planning and reporting cycles, directors should expect an integrated view rather than separate technology, finance, and risk presentations. Four decisions deserve explicit board attention:

• Inventory: require a current map of models, data, workflows, capabilities, owners, and dependencies.
• Evidence: approve thresholds for piloting, scaling, renewing, and stopping AI investments.
• Recoverability: define impairment triggers and escalation requirements before performance deteriorates.
• Transparency: align external disclosure with the judgments and risks used in internal decisions.

The strategic question is not how much the organization spends on AI, but how much durable value that spending creates and how confidently the board can prove it. With disciplined valuation, impairment review, disclosure, and governance, directors can distinguish durable AI intangible assets from expensed experimentation while preserving room to innovate.

Chris Winchester

Executive Managing Director, WGA Advisors

Chris Winchester is an Executive Managing Director with more than 30 years of cross-industry expertise in business strategy, operations, corporate strategy, and transformational change, with focus areas including M&A and acquisition integration, organizational development, digital and business transformation, supply chain optimization, business resilience, strategic value creation, AI, board governance, and private equity advisory. He holds an MBA from The Wharton School of the University of Pennsylvania and a bachelor’s degree in business from Pennsylvania State University.