An AI strategy must make investment choices

Name the investment the strategy would decline

A strategy becomes useful when it tells the company which attractive AI investment to decline. A hundred plausible applications can consume the same scarce experts without building a distinctive customer proposition. Decide which outcome merits sustained investment, then ask which projects contribute enough to earn those resources.

Consider a fictional industrial distributor seeking to make complex purchasing dependable for smaller customers. That proposition points toward product evidence and exception handling. A fictional bank seeking lower customer effort may prioritize case history and accountable remedies. Each choice changes the portfolio; neither is a mandate to automate everything.

Name the initiative the direction would postpone. If none qualifies, the strategy is too broad to allocate capital. Local efficiency projects can still be useful, but distinguish them from the commitments intended to improve the company’s competitive position. Those commitments need an accountable owner and a customer-and-economics argument.

Test what remains distinctive when models are shared

The cost and capability of models change much faster than many organizations can redesign their operations. Stanford's 2025 AI Index documents a historical decline in the inference price associated with a fixed benchmark performance level between November 2022 and October 2024. ¹ This is a benchmark-specific historical comparison, not a forecast of enterprise costs. Its strategic relevance is that an advantage based solely on access to a presently expensive component may be temporary.

Apply a rival-access test. If a competitor obtained the same model tomorrow, could it reproduce the customer outcome using information and relationships it can legitimately obtain? If so, model access offers little differentiation. If not, identify the permissioned evidence, trusted relationship or exception capability that makes the difference—and establish that customers value its effect. Data, processes and relationships are candidates for advantage, not proof of it.

Make the architecture reflect this distinction. Protect the records, evaluations and business permissions that give the service its identity. Keep model selection and execution infrastructure replaceable where the benefit warrants the cost. This is not a demand for universal portability. A deeply integrated supplier may offer a compelling advantage. It is a demand to price the commitment honestly: what becomes harder to change, which alternatives disappear and what evidence would justify accepting that dependence?

Buy delivery, evidence or an option deliberately

mayden’s Commitment Map separates three decisions. Delivery funding expands a demonstrated improvement. Evidence funding buys an answer to a consequential uncertainty. Option funding preserves a credible alternative without committing to the full service. These are management categories, not investment ratings. Each needs a different account of what the next expenditure will achieve.

Exhibit 01 gives the fictional distributor one change team and three competing proposals: expand its purchasing assistant, test the product-evidence approval path, or build a supplier-exit interface. If evidence rights and review ownership remain unresolved, fund the bounded evidence test before expansion. If repeatable service and capacity are demonstrated, delivery may take priority. Fund the exit option when a material dependence threatens a credible service and a small interface test preserves a useful alternative. No returns are assumed; the choice changes when the missing evidence arrives.

The Green Book discusses uncertainty and the value of incorporating flexibility into investment design. ² Applying that logic to private enterprise is mayden's interpretation. Staging is useful when a relatively small expenditure can change a much larger decision. It is less useful when successive pilots merely delay a commitment whose prerequisites are already understood. The board should ask what the next stage will reveal, what it costs and which decision becomes different when the answer arrives.

Set an actual review date before funding the next stage. Record the uncertainty, expected evidence and commitments that remain reversible. At that date, invest, extend for a specific unanswered question, or close the option. An experiment can finish successfully without justifying a product. Repeated extension without a changed question requires portfolio challenge.

Fund the complements that make the strategy executable

A strategic choice can fail even when the chosen technology performs well. The company may lack permission to use the necessary evidence, capacity to resolve exceptions or a route for bringing the service to customers. Research on the productivity J-curve models the role of complementary intangible investment. ³ It explains why measured returns can lag expenditure; it does not guarantee that a delayed programme will eventually succeed. Leaders need to identify the particular complement rather than invoke transformation as an explanation for weak results.

Build a dependency budget beside each major opportunity. It should show the operating expertise, data stewardship, legal work, integration and change capacity required to make the proposition real. Shared dependencies can justify common investment, but only when the relationship is specific. A general data platform should not inherit the claimed benefits of every future application. Assign each expected benefit once and describe the additional conditions needed for a downstream project to realize it.

This changes sequencing. If three promising services depend on the same unresolved evidence rights, another model pilot may add little information. The next investment belongs in resolving the rights and testing a usable data flow. Conversely, building an elaborate common platform before identifying a real service can consume capital without resolving demand. The sequence should follow the uncertainty that governs the business decision, with enough operational detail to expose the real prerequisite.

Treat supplier dependence as a strategic exposure

The CMA's foundation-model review identifies potential competition risks around critical inputs, distribution and interconnected partnerships. ⁴ These are regulatory assessments of market risks, not a prediction that a particular supplier will fail or exploit customers. They provide a reason to examine dependence at the level of the service. A second model account is of limited value if the evidence store, permissions and workflow remain inseparable from the first supplier's platform.

Supplier flexibility is an operating capability. In March 2026, the UK Competition and Markets Authority reported action on cloud egress fees and interoperability, while stating that further work was needed to help customers switch and use multiple providers. ⁸ For an AI investment, mayden would therefore price and test an exit route: portable data, reproducible evaluation, replaceable interfaces and a workable service during transition. A contractual promise to switch is weaker than a tested ability to do so.

For each material commitment, identify the change that would force a strategic decision: a price increase, an unavailable region, a change in contractual rights, retirement of a model or a persistent quality problem. Estimate the time and effort needed to move or narrow the service. Test a representative substitution when dependence is consequential. The objective is a credible choice under stress, rather than a diagram that declares the company vendor independent.

Dependence can still be rational. A supplier may offer security, distribution or operating capabilities that the firm could not economically reproduce. The decision should compare the present benefit with the expected cost of reduced flexibility, acknowledging uncertainty. Different workflows may warrant different choices. Procurement should therefore evaluate the relationship between contractual terms and the strategic proposition, rather than optimizing a unit price while the business unknowingly commits a critical capability.

Set boundaries that survive pressure for quick results

An investment direction should specify the outcomes the company will not pursue through unacceptable means. NIST's framework connects governance with context, measurement and management, while the OECD principles include accountability and human rights. ⁵⁶ These are normative references. mayden's application is to translate broad commitments into concrete limits on authority, evidence use and customer treatment before a high-profile programme faces delivery pressure.

A boundary should be operational enough to change a design. For example, a company may decide that a disputed eligibility decision requires an accessible human remedy and a retained evidentiary record. That choice affects cost, product design and the feasible pace of expansion. It also protects the credibility of the customer proposition. If the strategy depends on a service that cannot meet the boundary economically, leadership needs to change the proposition rather than quietly weaken the commitment.

Trade-offs should remain visible. More review may improve control while reducing responsiveness. Broader personalization may improve relevance while increasing evidence and privacy demands. A strategic document should name the important trade-offs and the executive who resolves them. The point is not to precompute every answer. It is to prevent the same unresolved disagreement from reappearing inside each project, where teams lack the authority to settle it coherently.

A strong direction must remain open to contrary evidence

The strongest objection to a clear strategic direction is that it can suppress discoveries. A company may reject an unfamiliar opportunity because it does not fit a narrative written before the technology changed. Reserve a bounded route for proposals that challenge the direction itself. Their task is to establish why a customer problem or economic possibility deserves reconsideration, not to claim immunity from evidence because the idea is disruptive.

Another objection is that staged investment favours measurable short-term efficiency over uncertain growth. The remedy is to match evidence to the decision. A new service may first require evidence of customer willingness to use or pay, rather than near-term margin. A capability investment may require evidence that several funded services depend on it. Government evaluation guidance stresses establishing how an intervention is expected to produce outcomes. ⁷ The company should state this causal logic in terms its leadership can challenge.

Review disconfirming evidence explicitly. Identify a plausible observation that would cause management to narrow, redirect or stop each major commitment. This protects teams from having to manufacture confidence to preserve sponsorship. It also distinguishes persistence from inertia. A strategic direction can remain stable while its implementation changes substantially, provided the customer advantage and the evidence supporting it still deserve investment.

Turn the strategic direction into three capital choices

Bring three choices to the board. Which customer outcome could improve the company’s position? Which capabilities deserve durable investment while technical choices remain open? What evidence earns the next commitment? Include the strongest alternative use of the same resources.

Translate those choices into the budget. Each major initiative needs a funding category, unresolved question, alternative, required complements and decision trigger. Finance separates committed from conditional spending; operations confirms capacity; commercial leadership establishes the route to a customer. Strategy has become actionable when these commitments agree.

Review what the portfolio is buying: repeatable delivery, decision-relevant evidence or a credible option. Stop financing ambiguity between them. A stable customer ambition can support substantial changes in technology and execution when the evidence points to a better allocation.