Capital under uncertaintyBetween false precision and unbounded learning

Requiring near-term cash return from every experiment can kill valuable information before it is available. Giving pilots an indefinite learning label can hide weak economics. Real-options theory helps frame commitment versus flexibility, but this article does not require complex option pricing. The practical question is what decision the spend keeps open, what information it buys, when the option expires and which irreversible commitment it delays.

Discovery-driven planning similarly makes assumptions and milestones visible. Reverse the claim: what customer behaviour, operational change, full cost and risk outcome must be true for the commitment to work? A bounded pilot tests the assumptions with the highest decision value. It should not merely demonstrate technical feasibility or repeat favourable cases.

The strongest rival is decentralised operating expense. Give units modest budgets, let them move quickly and hold them to normal P&L results. This avoids a slow central process and can fit low-consequence local use. It fails when reusable capability is duplicated, shared controls are unallocated, variable consumption is opaque, dependencies concentrate or one unit’s benefit shifts cost and loss elsewhere.

The capital-to-capture system can also fail. Finance invents precision, gates become procurement queues, sponsors relabel run cost as investment, options never expire or central allocations hide marginal economics. The ledger must preserve ranges, local speed and substantive judgement rather than reward a complete-looking spreadsheet.

A finance-grade decision therefore has three clocks. The evidence clock asks when the next material uncertainty can be resolved. The commitment clock asks when spend, dependency, workforce design or market promises become difficult to reverse. The capture clock asks when an operational change should reach a financial or service account. Treating those clocks as one produces predictable errors: a short pilot is expected to prove mature cash economics, or a long supplier commitment is approved on evidence that expires before the commitment does.

The governing unit is the capability and its decision path, not the model invoice. Context preparation, controls, integration, workflow change and recovery may dominate vendor consumption. Conversely, a reusable capability can support several later decisions even when the first consumer is small. Finance should trace which assets, obligations and observations survive each gate. That makes the record useful when leadership stops, narrows, transfers or expands the work, rather than only when the original forecast succeeds.

Six cost accountsFollow the capability through its lifecycle

Five value accountsDo not add unlike routes

Value captureThe capture covenant

A capture covenant states the value route, counterfactual, observation period, causal confidence, range, operational action, finance account, owner and review date. It also states the disconfirming test and stop threshold. The covenant is not a guarantee; it is the condition under which production scale remains funded.

For capacity, the action may be absorbing demand, reallocating work, reducing external spend or improving service. For risk, it may be lowering fraud loss while holding exposure and detection changes visible. For revenue, it may be launching a priced service with margin and adoption thresholds. For cash, it should reconcile to receipts, payments or budget. For learning, it should release a later decision before expiry.

Counterfactuals need care. Before-and-after comparisons can confuse the capability with process standardisation, management attention, demand or selection. Use a credible comparison, staged rollout, operational model or sensitivity range proportionate to the commitment. Causal confidence can remain a range, but finance should know which rival explanation threatens the case.

Complementary investment can delay measured productivity. The productivity J-curve research provides a reason to expect an initial period of intangible formation and organisational adjustment. It is not an excuse to roll forward weak evidence indefinitely. The ledger should show which capability asset is being formed, which consumers use it and when capture should become observable.

Set the review cadence from the evidence and exposure, not the annual budgeting calendar. Early discovery may need a short expiry because customer or technology conditions move quickly. Production value may require a longer observation window but tighter operating monitoring. A covenant can distinguish an early warning, a corrective threshold and a stop threshold. Crossing the first should trigger investigation; crossing the last should change funding unless an authorised decision records new evidence and the cost of delay.

Ownership must be joint but not blurred. The operational sponsor owns the workflow action and customer or service result. Product and technology owners carry capability performance, consumption and dependency evidence. Risk owners challenge exposure and recovery. Finance owns reconciliation, allocation integrity and the connection to capital decisions. A committee is not an owner. Each covenant field needs one accountable role and an escalation path for unresolved evidence.

Ranges should narrow for a reason. Replace an assumption with a measured distribution, contract term, observed adoption rate or tested recovery time. Do not narrow a forecast merely because a gate requires a precise number. When evidence remains weak, show sensitivity across demand, cost, loss and delay. The decision can still proceed if downside is bounded and reversible, but scale should not inherit discovery-level confidence without stating the risk.

Portfolio economicsShared cost, variable consumption and supplier commitment

Shared-platform double counting occurs when several cases claim the full enabling benefit and none carries the common cost. Maintain the capability as its own account with intended consumers, allocation rule and residual value. Consumer cases carry marginal onboarding and run cost, plus a transparent share for decision support. Allocation should aid accountability, not disguise the economics of the next use.

Variable AI consumption needs unit drivers: requests, tokens, tool calls, reviewed cases, resolved outcomes or customer journeys. Average cost can hide a costly exception tail. Marginal cost can change when capacity, rate limits or review thresholds step. Finance, product and platform teams should agree which unit follows value and which follows spend.

Supplier discounts can exchange current price for minimum spend, concentration and exit exposure. Compare the present saving with unused commitment risk, switching assets, termination terms and reduced bargaining flexibility. A vendor commitment may still be rational. The ledger should show the dependency rather than call the discount unconditional value.

Controls and resilience are part of economic design. NIST lifecycle governance and the Basel operational-resilience principles identify roles, monitoring, dependencies, response and recovery that create real cost and loss limits. Apply them within scope and proportion. Removing their cost from the investment case would compare an uncontrolled aspiration with a production capability.

Portfolio reporting should distinguish formation health from consumer value. A shared context or evaluation service can be delivered on time while consumers fail to capture value; a valuable consumer can also survive despite a weak shared platform. Show adoption, reliability, variable consumption, allocation and exit exposure at the capability level, then show the causal value route for each consumer. This prevents a successful local case from certifying every platform claim.

Allocation has two purposes that should not be confused. Stewardship allocation shows where common cost and capacity are consumed. Decision allocation helps compare the next consumer, supplier tier or feature. The first may use stable policy for planning; the second must expose avoidable and incremental cost. A business case should not reject a beneficial new use because of an arbitrary share of sunk formation cost, nor approve it by pretending shared services are free.

Commitment is a portfolio property as well as a contract field. Several small renewals can create concentration through common models, scarce skills, context formats or review processes. Aggregate minimums, correlated failure, migration demand and negotiating leverage across capabilities. Finance should ask whether the portfolio can change direction within its stated time and loss tolerance, not merely whether each purchase met its own approval threshold.

Negative controlsCases that expose accounting fiction

Interactive instrumentCFO AI Capital and Accountability Console

The Console separates stage, costs, value evidence, capture, supplier commitment and stop logic. It returns a funding route and conditions, not one ROI or accounting conclusion.

The CFO operating agendaBuy learning early, demand capture before scale

Create one capital-to-capture record for each material capability, not one ROI for the portfolio. Link consumer cases to shared capability accounts. Preserve cost and value ranges, evidence dates, causal rivals and owners. Use the commitment ladder to decide what proof the next release of capital requires.

Give discovery a priced question and expiry. Give reusable capability named consumers and an allocation rule. Give production a capture covenant and finance reconciliation. Give hard supplier, workforce or market commitments a tested downside and exit case. Keep accounting policy review visible but outside the economic route.

Review forward economics. Historical spend can reveal what was learned and what residual asset remains. It cannot make a weak next period attractive. Stop, sell, narrow or redeploy when the future mechanism fails, and account for recovery rather than hiding it.

Finance discipline should improve speed by matching evidence to the decision. Small reversible learning does not need scale-grade precision. Irreversible production should not rely on pilot optimism. The CFO adds value by making the transition explicit, so capital buys information when uncertainty is high and earns capture when the organisation chooses to commit.

Run the agenda through existing planning and performance rhythms, but preserve stage-specific decisions. Monthly operating reviews can reconcile consumption and covenant signals. Quarterly portfolio reviews can challenge shared allocation, correlated dependency and capability reuse. Capital releases should cite the evidence gained since the last gate and the commitment created by the next one. This creates continuity without forcing every decision into one annual forecast.

Publish internal decision receipts that state route, range, conditions, owner, review date and accounting-review status. Receipts make exceptions legible and allow later comparison between expected and realised mechanisms. They also expose repeated optimism, unclosed pilots and benefits claimed by several sponsors. Accountability improves when the record shows why a reasonable decision was made under uncertainty, not only whether the eventual outcome was favourable.

The management view should make five questions easy to answer. What uncertainty is the next pound buying down? Which costs and dependencies become unavoidable after the next gate? Through which account will value be captured? What observation would reduce or end funding? Which reporting questions remain with qualified policy and audit specialists? If those answers cannot be reconciled from the ledger, apparent portfolio precision is not decision control.

Use post-decision review to improve the system, not to reward confident forecasts. Compare the expected mechanism, range and timing with what occurred, then identify whether the variance came from evidence, execution, demand, cost, dependency or capture. Feed those observations into later ranges and gate design. A well-governed option can end with no scaled service and still be a sound investment if it avoided a larger mistake. A favourable outcome does not excuse an uncontrolled decision path.