Frame the value question and engagement boundary

The value hypothesis is an investment argument, not a technology description. It must show how the customer-service copilot changes a named decision or workflow, who receives the benefit, and which baseline permits comparison. Record purpose, affected process, beneficiary, counterfactual baseline. If the baseline is unstable, the first recommendation concerns measurement or process repair rather than model acquisition. The scope record for customer-service copilot must distinguish the asserted need from observations about purpose, affected process, beneficiary, counterfactual baseline.

Portfolio diagnosis tests demand from three directions: strategic necessity, operational friction and credible user need. For the customer-service copilot, interview the process owner, frontline users and affected parties separately; their incentives differ. Reconcile observed cycle time and error patterns with interview claims, and mark unsupported benefit estimates as hypotheses rather than silently converting them into business-case facts. A contrary explanation for customer-service copilot should remain visible until evidence tied to value hypothesis eliminates it.

Compare the following genuinely fundable paths: automate broadly; run a bounded pilot; defer pending process repair. Estimate total lifecycle cost, dependency concentration, time to first useful evidence, reversibility and downside exposure. A small experiment can be valuable when it resolves a decisive uncertainty; it is waste when success criteria are vague. The justified position here is to run a bounded pilot with explicit non-goals, conditional on a named evidence gate. Sensitivity analysis must show whether the recommendation to run a bounded pilot with explicit non-goals remains justified after a plausible change in cost, exposure or readiness.

Stage-gate funding separates permission to learn from permission to scale. Gate one verifies process ownership and data fitness; gate two verifies usable performance and affected-party safeguards; gate three verifies adoption, economics and operating capacity. The customer-service copilot advances only when evidence meets the pre-agreed threshold, not when sunk cost or executive visibility makes stopping uncomfortable. The accountable owner for customer-service copilot signs the next decision, while evidence owners sign only what they observed.

Portfolio balance is tested under constraint, not by adding project scores. Reserve capacity for shared data, evaluation and change support, then examine whether several initiatives depend on the same scarce team or vendor. The value hypothesis should expose correlated failure: three individually attractive projects may create one fragile portfolio when their assumptions, infrastructure or regulatory exposure are identical. A failed threshold changes action: narrow, contain, redesign or stop customer-service copilot; it never becomes a decorative alert.

A portfolio review for customer-service copilot compares forecast with realized adoption, unit cost and outcome movement. It also checks whether another initiative produced the same benefit more cheaply. Record attribution limits and avoid treating activity volume as value. Use purpose, affected process, beneficiary, counterfactual baseline to explain which observation supports the conclusion and which remains absent; the customer-service copilot record must not merge the two.

The decision memo states the selected path, rejected alternatives, assumptions, funding tranche and next decision date. For customer-service copilot, include one benefit indicator, one harm or exclusion indicator, one adoption indicator and one cost indicator. Assign each indicator an owner and a response rule so reporting changes the portfolio rather than merely describing it. The handoff package links value hypothesis to one inspectable record, one response rule and one review date.

When a portfolio assumption fails, identify which projects share it. For customer-service copilot, update dependency and concentration maps before reallocating funds; otherwise the organization may repeat the same error across several apparently unrelated investments. Re-test value hypothesis after the first operating cycle and compare the observed constraint with the one assumed when recommending run a bounded pilot with explicit non-goals.

Closure preserves learning. Archive the decision basis, evidence, unused options and reason for stopping or scaling customer-service copilot. Feed the lesson into the next value hypothesis review rather than rewarding teams only for launches. The closure reviewer can reopen the customer-service copilot judgment when benefit depends on adoption and process change; this condition is recorded beside the owner and evidence date.

Limit the claim to this portfolio constraint: benefit depends on adoption and process change. Do not infer realized value from prototype quality, user excitement or vendor benchmarks. If evidence remains mixed, narrow the commitment, acquire the discriminating information and preserve the option to stop. This is disciplined capital allocation, not reluctance to innovate. Because benefit depends on adoption and process change, the final wording states a bounded conclusion and the condition that would reverse it.