Direction under uncertaintyA thesis is a coordinating claim, not a corporate belief

Strategy requires choice and fit across activities. Porter’s formulation is useful because an AI initiative does not become strategic merely by using a capable technology. It matters when it reinforces a distinct value proposition and a system of activities that competitors cannot reproduce by buying the same component. If access to a model becomes broadly available, advantage must sit elsewhere: customer access, trusted distribution, proprietary context rights, an embedded workflow, a physical network, a difficult operating integration or a faster learning loop.

The strong rival is emergent strategy. Mintzberg and Waters describe deliberate and emergent strategy as ends of a continuum rather than mutually exclusive camps. Under high uncertainty, many cheap local experiments can reveal customer demand, operational constraints and unexpected combinations. That is defensible. It becomes costly when scattered commitments produce incompatible platforms, duplicated controls, fragmented customer journeys, supplier lock-in or irreversible workforce changes before leadership says what kind of company those choices should create.

The answer is not to centralise every experiment. The CEO thesis supplies direction and common boundaries; local teams supply variation and evidence. Experiments may inherit the thesis, test an assumption inside it, or explicitly challenge it. Contradictory results are not disloyal. They are strategic information. What leadership must prevent is an experiment silently creating enterprise infrastructure or customer promises that narrow future choice without an acknowledged commitment.

March’s exploration and exploitation tension sharpens the resource question. The company needs enough exploration to discover a changed opportunity and enough exploitation to compound what works. Capital allocation, executive attention and shared platforms turn the rhetorical balance into actual strategy. Bower and Gilbert’s resource-allocation lens warns that realised strategy appears in funded positions, approved vendors, operating targets and leadership time, not only in the annual narrative.

Thesis syntaxWrite one enterprise sentence that exposes its assumptions

A useful draft can follow a demanding syntax: because a defined form of intelligence has changed within a stated horizon, the firm will concentrate on a chosen customer decision or workflow; it expects a specified value mechanism; it will compound named distribution, context, rights or learning advantages; it will commit a bounded set of capabilities, capital and role changes; and it will revise when a stated contradiction crosses a dated threshold. The sentence will be long. Its purpose is diagnostic. Any vague clause points to a claim the leadership team has not resolved.

The CEO can then separate the sentence into the ten register entries without losing the connections. This matters because functions naturally translate the thesis into their own nouns. Technology sees architecture, finance sees return, HR sees roles, product sees customers and risk sees exposure. The syntax keeps them attached to one mechanism. A context platform is not strategic if it does not serve the chosen arena. A cost case is not complete if the stakeholder settlement changes adoption. A risk limit is not effective if the adaptation rule allows dependent commitments to continue.

Use ranges where the evidence supports ranges. A horizon can be “within the next two planning cycles” rather than an invented month. A value mechanism can carry a downside and a break-even adoption boundary. The register should preserve which statements are observation, theory, modelling and management judgement. Confidence language is useful inside each claim, but aggregating it into one thesis confidence score would hide which dependency can overturn the whole system.

The enterprise thesis also needs a short list of non-choices. These may include refusing to build a proprietary foundation model, declining high-consequence autonomous action until takeover evidence exists, avoiding shared capital for disconnected personal-productivity tools, or postponing a market entry that lacks distribution rights. A non-choice is not permanent. It is the current consequence of the thesis and should change only when the contradiction register supplies a reason.

Claims 1–4From structural change to chosen advantage

Activity-system fitTest whether the claims reinforce one another

Read the first four claims as a causal set. If the structural change is cheaper document generation but the arena is a trust-sensitive advisory decision, cost may not be the binding value mechanism. If the chosen advantage is proprietary context but customers will not grant reuse rights, the feedback loop is assumed rather than available. If the arena spans every function, no coherent activity system can guide capital. The register should flag such incompatibilities before executives negotiate them into vague wording.

Fit also appears across time. A reversible probe can rely on manual context preparation, but a scale thesis may require a governed context service. A temporary model lead can test demand, but it cannot remain the long-term advantage claim. A workforce settlement can be staged while task evidence develops, but customer promises should not expand faster than takeover and challenge capacity. Each stage must inherit the logic of the next without pretending the next commitment is already authorised.

Trade-offs make the fit visible. Choosing a deeply embedded workflow may sacrifice rapid horizontal expansion. Preserving supplier optionality may raise current integration cost. Building trusted distribution may require slower evidence and recourse design. Protecting apprenticeship may retain some independent practice that a narrow automation case would remove. These are not defects to edit out of a board paper. They are the costs of the chosen position and should be compared with the advantages they make possible.

Claims 5–7Translate direction into capability, capital and settlement

Claims 8–10Bound loss, protect contradiction and specify adaptation

Negative controlsWhat the register must do when the story breaks

Worked thesisFrom document cost to recurring decision service

Consider a business-services firm whose first claim is that generative AI will reduce document-production cost. It funds reusable context, document evaluation and a staged delivery assistant. Early evidence shows modest cost improvement after verification, but clients repeatedly ask for the reasoning context, ongoing monitoring and decision follow-up. The local experiments reveal willingness to pay for a recurring decision service rather than cheaper documents.

The contradiction register prevents leadership from classifying the requests as feature demand inside the old thesis. The strategic arena moves from document workflow to recurring client decisions. The value mechanism moves from labour substitution to service growth and retention. The advantage moves from model quality to trusted distribution, embedded workflow, proprietary context rights and outcome feedback. Capital shifts from document automation alone to product, account integration, evaluation and service operations.

Reframing still needs evidence. Customer requests may reflect novelty or sales attention. A staged product test should compare willingness to pay, adoption, delivery economics and outcome quality. Workforce design changes too: experts may spend less time producing routine documents and more time framing decisions, challenging evidence and governing the feedback loop. The thesis becomes stronger because the contradiction changes both the advantage and the commitments, not because executives tell a better story.

Contradiction protocolGive inconvenient evidence a route to resource allocation

A register becomes ceremonial if contradictions are merely logged. Each material signal needs an owner who is not rewarded only for delivering the current thesis, a forum with authority, a review date and a pre-agreed consequence. The consequence might release a bounded probe, freeze a supplier commitment, move product capital, revise a workforce plan or withdraw a customer promise. Without a resource effect, leaders can acknowledge the evidence while realised strategy continues unchanged.

Classify contradictions by what they attack. A boundary contradiction shows that the observed evidence covers a different population or workflow. A mechanism contradiction shows that the claimed cause does not explain the outcome. An advantage contradiction shows that a rival can imitate or a supplier controls the learning loop. A commitment contradiction shows that capital, talent or stakeholder terms cannot support the direction. An adaptation contradiction shows that the organisation cannot reverse what the thesis calls reversible. Classification identifies which claims must be recompiled.

Weak signals should not trigger theatrical strategy changes. The adaptation rule can require replication, a threshold across customers or a bounded comparison. Yet leadership must avoid raising the evidence bar only after a threatening result appears. Pre-registering the observation, date and decision effect reduces that temptation. It also lets teams design experiments that are informative enough to matter rather than producing another favourable demonstration.

When the thesis changes, communicate the reason in the same architecture. State which structural assumption, mechanism or advantage changed; which commitments continue; which are frozen; how stakeholders are affected; and what the new contradiction would be. This maintains direction during revision. The organisation learns that strategic conviction and intellectual honesty are compatible because both operate through visible claims and decisions.

Interactive instrumentCEO AI Thesis Compiler

The Compiler stores an owner, dated evidence, dependency, contradiction and adaptation note for each of the ten claims, then tests cross-thesis signals. It returns a decision route and next evidence, never a confidence or maturity score.

CEO operating cadenceUse conviction to allocate, and contradiction to adapt

The CEO should own the integrated thesis even when executives own individual claims. Strategy can prepare the architecture, technology can own system capability, finance can reconcile economics, HR can own workforce transition and risk can challenge boundaries. None should optimise its claim independently. A cheaper platform that traps context may weaken advantage. A faster workforce change without a learning path may weaken adoption and control. A strong customer signal may justify moving capital before every technical preference is settled.

Review cadence should follow the three clocks of capability, organisational change and governance rather than one quarterly rhythm. Fast technical observations may change a probe. Slower customer adoption may govern a product stage. Capital, role and risk-appetite changes need their own authority. The contradiction register links these clocks by stating which observation reaches which forum and which dependent commitment pauses.

The board receives the thesis as management’s falsifiable enterprise position and applies the proof questions from Paper 061. Directors should see the arena, mechanism, advantage, commitments, stakeholder settlement, risk boundary and adaptation rule. They do not need every local experiment. They do need material contradictions, changed exposure and resource movements that alter the original commitment.

A useful thesis makes saying no easier and changing course more credible. It refuses generic projects that borrow the AI label without reinforcing the chosen system. It funds shared capabilities that improve several aligned bets. It permits bounded probes that may overturn the claim. Leadership earns the right to persist not by sounding certain, but by specifying the evidence under which persistence remains rational and the threshold at which it will stop.