Conviction and the Pruning Problem

Every major organisation has now deployed AI. They run more scenarios than ever before. Strategy decks that used to take months take minutes. And somehow, decisions are getting slower. That is the visible half of the problem. The other half arrived quietly: a probe that once required a product squad can now be built by one person over a weekend. Variants spawn in parallel. Interfaces mock instantly. The build wall is falling.

Two pathologies, one root cause. AI solved the information problem and collapsed the build cost. What neither development solved, and what both made dramatically worse, is the discipline to decide. To commit. And to kill. Those two gaps, the commitment problem and the pruning problem, are where the next decade of competitive advantage lives or dies.

The bottleneck is no longer production. The bottleneck is judgment. AI lets firms build more than they can evaluate and confuse motion with learning at unprecedented speed.

The Conviction Deficit

For three decades, corporate strategy absorbed one idea from finance: optionality has value. Keep options open. Defer commitment until uncertainty resolves. Dixit and Pindyck gave the discipline its rigorous form in Investment Under Uncertainty, and a generation of strategists internalised the conclusion: waiting is rarely free, but it usually buys information, and information reduces uncertainty.

AI demolished the silent assumption underneath that logic. Real options theory breaks because waiting no longer yields decision-relevant information. When you can run a thousand scenarios before lunch, the marginal information value of additional analysis approaches zero. The uncertainty that remains is not the resolvable kind. It is irreducible ambiguity. No model will tell you what the geopolitical environment looks like in eighteen months. Meanwhile, the cost of holding the option open is climbing. Competitive windows compress. First-mover positions get occupied while your team runs its third round of sensitivity analysis.

So the binding constraint on value creation moves. It is no longer information, insight, or even decision speed. It is conviction: the willingness and capacity to commit resources under irreducible ambiguity. The scarce resource is the willingness to commit when the models disagree. We call that Conviction Capital, and it is becoming the next source of alpha for anyone who can act while everyone else is still running one more scenario.

Three Measurable Constructs

Conviction is not a mood. It can be priced. Three constructs make what was previously unmeasurable tractable.

I. The Conviction Premium. The value captured by committing before market consensus, minus the expected cost of being wrong. When AI compresses the gap between early information and consensus timing, but strategic windows still reward first-movers, this premium becomes the dominant source of alpha. It is structurally analogous to the exercise premium on an American option versus a European one: the value of being able to act before expiry rather than waiting for it.

II. Conviction Volatility. The volatility of organisational commitment itself. Some organisations commit and hold. Others oscillate between commitment and reversal, initiative and retreat. High conviction volatility is purely destructive. An organisation that commits, reverses, then recommits pays the exercise cost multiple times while capturing the value zero times. It is measurable: track the frequency of strategic reversals, the half-life of capital allocation decisions, the ratio of projects initiated to projects completed. This belongs in enterprise valuation as a discount factor. Analysts do not currently measure it. They should.

III. The Conviction Surface. Most organisations show high conviction on small, short-term decisions and low conviction on large, long-horizon ones. That is a steep surface. Sovereign wealth funds, founder-led firms, and long-horizon family conglomerates consistently outperform because they have flatter surfaces. They commit to thirty-year positions with the same decisiveness they apply to weekly operational calls. That is not a governance quirk. It is measurable competitive advantage.

The deficit cannot be fixed by telling leaders to be more decisive. That is a slogan, not a fix. The fix is architectural: separate the Intelligence Function from the Commitment Function and install a one-way valve between them. The Intelligence Function runs continuously, modelling scenarios and synthesising signals, and it never decides. The Commitment Function converts whatever the intelligence surface shows into a binding resource allocation. Not a recommendation. A commitment. The container for that conversion is a seventy-two-hour Decision Sprint. Hours 0 to 24: frame and dump, reframing the decision as a binary or small-set choice and surfacing all existing intelligence with no new primary research. Hours 24 to 48: stress test, with two teams arguing opposing positions and each naming the single condition under which it fails. Hours 48 to 72: commit, documenting what you are doing, what you are explicitly not doing, and what must be true for you to reverse the decision. That third element is what prevents conviction volatility.

The Pruning Problem

When build cost collapses, the new pathology is what you refuse to kill. The cleanest illustration is biological. In 2010, researchers placed oat flakes on a surface in positions corresponding to cities around Tokyo, introduced a brainless single-celled organism called slime mold, and watched it grow. At first it explored broadly. Then it reinforced the channels that carried useful flow and let the weak channels disappear. The resulting network resembled the Tokyo rail system in cost, efficiency, and fault tolerance.

The lesson is not that nature is clever. The lesson is mechanical. The organism does three things at once: it explores cheaply, reinforces corridors that carry signal, and prunes corridors that do not. There is no strategy offsite, no committee, no annual planning cycle, no executive defending last quarter's mistake. The intelligence is not inside a brain. It is inside the feedback loop. That loop is what most companies cannot run, and AI has just made the failure mode significantly worse.

When probes were expensive, the cost gradient did the discipline for you. Weak ideas died on their own. When probes become cheap, weak ideas no longer die. They linger.

They stay in the Slack channel, the roadmap appendix, the innovation tracker. No one wants to kill them because each one seems inexpensive in isolation. But the cost has not disappeared. It has moved, from capital expenditure to attention, from engineering hours to coordination overhead, from budget line to decision bandwidth. This is the AI-era pathology: organisations will accumulate zombie probes. Hundreds of prototypes. Dozens of half-live workflows. Many plausible demos. Few hard decisions. They will call this a portfolio. It will actually be a landfill. The conclusion is uncomfortable but clean. In the AI era, strategy is not a plan. It is a pruning system.

Four Kill Rules

If AI removes the natural constraint on experimentation, firms have to construct an artificial one. The question is no longer whether you can build something. Increasingly the answer is yes. The better question is what must die for this to deserve continued life. Four rules force the answer.

1. The Option Value Floor. Kill a probe when the value of keeping the option alive falls below the cost of carrying it to the next decision point. A probe is not valuable because it exists. It is valuable because continuing it preserves meaningful upside under uncertainty. Once that option value drops below the cost of attention, governance, compliance, or executive bandwidth, it is no longer an option. It is a liability.

2. The Execution Cost Gate. Kill a probe when the expected cost of reaching the next real signal exceeds the expected value at current signal strength. AI collapses build cost. It does not collapse customer access, distribution, integration, legal review, or executive sponsorship. A prototype may be free. Progress is not.

3. The Slot Opportunity Cost. Kill a probe when the next-best probe deserves the slot more. The firm has finite attention. A probe does not compete against nothing. It competes against every other probe that could occupy the same scarce slot. The question is not whether this is somewhat promising. The question is whether it is more promising than the marginal alternative.

4. The Volatility Collapse Rule. Kill or promote a probe when uncertainty has collapsed. Options are valuable because the future is uncertain. Once uncertainty disappears, the probe should stop being a probe. If the signal is clearly positive, promote it to operations. If clearly negative, shut it down. What should not happen is the common corporate middle state, where the probe stays alive because nobody has forced a classification. A probe with no remaining uncertainty is not optionality. It is indecision.

The hard part is not methodology. It is political economy. The probe that should die has a sponsor. The dashboard that should decide is contested. The metric that looked clean at launch becomes negotiable once careers attach to it. Enforcing the rules requires a defended definition of what counts as evidence, an authority empowered to act on it, and a board that treats visible failure as the cost of signal generation rather than as embarrassment.

What to measure, and the Q4 board question

The framework reduces to four metrics. They are the instrument panel for both pathologies: conviction and pruning, commitment and kill.

The board question for the quarter is therefore simple to ask and uncomfortable to answer. Does this AI investment help us decide, or help us defer? Does it help us prune, or help us accumulate? The slime mold solved this problem six hundred million years ago. It has no ego, no sunk cost, no internal champion. It prunes because its body forces the issue. Companies have to engineer what biology gave the organism for free.

The firms that learn to prune will compound. The firms that cannot will collect prototypes and call it transformation.