Open Source Executive Briefing CW37: What Should Executives Assess Before Funding an AI Opportunity?

An AI opportunity assessment should test the value of a proposed change and the enterprise conditions needed to make it work. It should also use AI itself to bring credible evidence forward, before leaders commit to expensive transformation.

Some alarm-bells should ring when an organization declares that it is not ready. Sometimes an essential safeguard or capability is genuinely missing. Sometimes the proposed opportunity is being judged against an operating model treated as permanent, even though changing that model is where the value might lie.

The executive task is therefore larger than checking readiness. It is to distinguish a weak opportunity from a changeable organizational constraint, then decide how cheaply and responsibly that distinction can be tested. AI belongs on both sides of this decision: as a potential source of business value and as a tool for investigating the case.

Keep opportunity evidence and enterprise conditions separate

Organizational maturity and opportunity quality answer different questions. A sophisticated platform does not establish that a proposed workflow change is useful. An immature enterprise does not automatically invalidate a bounded opportunity. Both judgments require a more specific account of the work.

A useful AI opportunity assessment therefore asks two separate questions: is the opportunity itself weak, or is a changeable enterprise condition blocking a worthwhile test?

Opportunity evidence concerns the problem, its connection to a consequential business objective, the current baseline, the proposed value mechanism, and what would demonstrate financial viability, technological feasibility and internal desirability. The assessment also needs an accountable owner and a clear statement of what remains unknown.

Enterprise conditions concern the data, permissions, governance, skills, architecture and decision rights that affect the proposed next step. These conditions should be examined as actual constraints, not condensed into one score. Business strategy supplies direction; evidence establishes which changes deserve commitment.

For example, a workflow that depends on several sequential handoffs may have been designed around the cost of accessing expertise. Its feasibility should not be judged solely by whether AI fits neatly into each existing handoff. The assessment should also test whether some handoffs remain necessary. That is a design hypothesis, not permission to remove accountability.

Use AI opportunity assessment to bring evidence forward

OSC recommends using AI to organize the investigation before commissioning a large transformation programme. It can help cluster proposed opportunities, flag potential duplicates, compare assumptions, prepare prototypes and identify questions for users or specialists. Those outputs should make it easier to conduct a real test, rather than simply produce a more persuasive investment presentation.

The research supports opportunity, but also restraint. In The Cybernetic Teammate, published in June 2026, professionals at Procter & Gamble working individually with AI matched teams without AI on the study’s innovation-performance measure. The authors found that AI primarily improved idea generation, while human judgment retained value in selection. This is evidence about a particular innovation setting, not proof that AI can validate commercial demand independently.

AI is not free of bias. NIST identifies human, systemic and computational sources of AI bias. Removing a sponsor’s name from an assessment may reduce one influence, but it does not make the model’s criteria, inputs or conclusions neutral. AI-assisted challenge should expose assumptions to scrutiny, not replace one authority with another.

The Jagged Technological Frontier experiment provides a related warning: AI improved performance on tasks within the tested model’s capabilities, but reduced correctness on a task outside them. Those historical task boundaries are not a map of today’s models. The relevant lesson is to test the quality of AI assistance in the assessment itself.

A prototype is a testing instrument, not the evidence by itself

An AI-generated prototype can make an opportunity concrete enough for a critical group to challenge. Instead of asking whether employees like an abstract idea, let them attempt a representative task. Observe where they succeed, what they correct, which exceptions break the approach and whether the proposed workflow is preferable to the current one.

Within an AI opportunity assessment, the prototype is the testing instrument; evidence comes from what happens when it meets people, tasks and operational conditions. A convincing demonstration can support further investigation without proving sustained adoption, production reliability or financial return. Each test should make clear which question it can answer and which questions remain open.

Lower preparation effort can make a broader set of tests practical. It does not eliminate the effort of recruiting relevant participants, validating data or reviewing results. The objective is more useful learning per unit of effort, not more prototypes competing for the same attention.

Judge access across the experiment, not seat by seat

Requiring a proven return before granting any access can make the evidence impossible to obtain. The better question is whether the cost and exposure of a bounded trial justify rejecting it, given what could be learned. The assessment process itself also consumes time and attention; scrutinizing a small experiment should not become more expensive than conducting it.

A trial can be economically worthwhile even when only some participants discover valuable applications. The successful uses must, however, be evaluated against the total cost of the trial: all licenses, employee time, support and relevant implementation or risk-management work. A strong return for one successful seat does not establish a positive return for the whole group.

This is an argument for proportionate access with a purpose and review point, not indefinite unused subscriptions. Microsoft’s Copilot Adoption Playbook recommends intentional seat assignment, specific use cases and support for new work habits. That is vendor guidance on adoption design, not independent proof of return.

Put the larger commitment boundary after the first test

The boundary should distinguish permission to learn from permission to transform. Initial exploration can be small and reversible; a larger commitment should follow the first credible test and identify what the evidence now justifies. This develops the AI Value Hypothesis Gate without treating every experiment as a miniature transformation programme.

Baseline controls still apply before testing. Approved tools, appropriate data, clear permissions and human review are conditions of a responsible experiment, not rewards for a successful business case. NIST’s AI Risk Management Framework treats governance as continuous and cross-cutting. Moving the investment gate after learning must not mean moving all risk assessment after exposure.

Nor can AI resolve a leadership preference for a passion project. Decision-makers should agree beforehand what result would cause a stop, redesign or further test. Where sponsorship overrides negative evidence, that override should be explicit. Faster analysis creates little decision value when the organization is unwilling to act on its findings.

Let the diagnosis determine the next commitment

The portfolio-to-enterprise feedback loop becomes useful when assessment results lead to different actions. The following distinctions are decision prompts, not a replacement maturity score.

After the first test, a continuing initiative should carry an owner for the next commitment, a defined evidence requirement and a review point. A failed test should leave reusable learning. Neither outcome needs to become another open-ended pilot.

Leadership Question

Are we rejecting a weak AI opportunity, or protecting an operating model that AI could help us redesign?

A disciplined AI opportunity assessment uses AI to test assumptions earlier, keeps opportunity evidence distinct from enterprise conditions, and preserves human accountability for the commitment. AI can help leaders discover that a constraint is no longer fixed. Evidence must still establish whether changing it is worthwhile.