More information is not always a better decision

A research request becomes a decision problem when an institution can name an action that could change after the result arrives. That action may be a commitment, a staged trial, a withdrawal, a redesign or a deliberate decision to wait. A report title is not an action. Neither is a general instruction to understand a market better. Before discussing a study design, the commissioning team should state what the result could permit, prevent or alter. This gives researchers a target and gives the committee a basis for deciding whether the proposed evidence is worth buying.

Consider a committee facing an uncertain opportunity. It can commit at scale, preserve a smaller position or hold its resources. The uncertain state affects the payoffs of each action. A proposed signal can change the committee's view of that state. The essential comparison is between the best feasible action with current information and the conditional actions that would be selected after each possible signal. The signal creates decision value through those action changes. A more detailed description of the uncertainty is useful only to the extent that it improves a relevant choice or serves another explicitly declared institutional purpose.

The practical difficulty is that organisations often commission a study before agreeing on the menu of actions. A report then arrives to an audience that cannot change the design, lacks the authority to change the contract, or has already spent the budget. The information may be intellectually useful while its commissioning logic has failed. A decision charter should therefore describe the action owner as well as the research question. A committee that can recommend a change but cannot authorise it must identify the subsequent approval path and test whether that path can complete before the opportunity expires.

An information charter also needs a counterfactual. The relevant alternative to commissioning a study is not ignorance in the abstract. It is the action the institution would take using the information it already has, including its existing uncertainty, operational safeguards and available options. If the team would already stage the investment, a study must improve upon staging. If a binding condition already requires a safety investigation, a voluntary purchase calculation cannot cancel that requirement. If the team would hold, the study must explain why its possible findings could make a feasible alternative preferable.

This makes the timing of the charter consequential. Writing it after the results are known invites a retrospective justification of almost any study. An unfavourable result can be called a useful warning; a favourable result can be called confirmation; an ambiguous result can be called a reason for another report. Each interpretation may be reasonable in a particular context, but without a recorded prior decision and a conditional response rule there is no stable way to distinguish a useful commission from a research process that continually changes its stated purpose.

The charter should preserve the distinction between information and intervention. A test may reveal that an operating design is weak without making that design easier to change. A survey may identify demand without creating a supply capability. A technical pilot may demonstrate a component while leaving its integration unresolved. The resulting decision value depends on the responses that the institution can implement, rather than the responses an analyst can imagine. This is why the paper's frozen action menus are explicit. Their admissibility is assumed for illustration and remains an independent implementation question.

The same discipline applies to an apparent decision to do nothing. Holding resources may preserve flexibility but can also sacrifice an expiring option, a learning opportunity or an obligation to maintain service. Its payoff belongs in the same perimeter as the active choices. A zero hold payoff is a convenient illustrative baseline, not a universal claim that inaction has no consequences. If holding changes obligations, relationships or future access, the payoff matrix must represent those consequences or explain why they are handled outside the monetary comparison.

An actionable research question is narrower than a comprehensive account of a business. The committee may need to know whether a specified operating condition is likely enough to justify a commitment. It may not need a general industry forecast, an exhaustive technology history or a complete explanation of every uncertainty. Research scope should follow the decision's sensitivity. A narrow study that resolves a binding choice can be more useful than a broad study that improves understanding of variables that do not alter the selected action. Scope discipline also makes the claim support more reviewable.

This does not imply that every worthwhile research activity must have an immediate monetary return. Institutions may commission foundational science, public evidence, capability development or long-term exploration. Those purposes should be named directly. A short-horizon expected monetary value calculation is poorly suited to an exploratory programme whose action set is expected to change. The error is to attach a numerical purchase rule to one purpose while relying on a different purpose to justify the expense. A mixed commission should separate its decision-support component from its broader public or exploratory objectives.

In the illustrative useful-signal case, the decision is deliberately simple. A binary state affects the payoffs of Commit, Stage and Hold. A binary signal has supplied conditional likelihoods and a stated purchase cost. The canonical engine chooses the action with the greatest expected payoff before information, calculates the posterior in each possible branch and chooses again. This stripped-down structure makes the mechanism visible. It does not estimate an actual opportunity's likelihood, validate a study instrument, or establish that a committee has the authority to implement the selected branch.

The value comparison should be recorded before the study is sold as persuasive evidence. A supplier can describe a highly accurate instrument while leaving the client's response rule unspecified. A sponsor can request a detailed forecast while assuming a project must proceed regardless. A committee can welcome a useful report and still have no intention of changing its position. These are commissioning problems that accuracy alone cannot repair. The charter should ask what adverse finding would stop or modify the commitment and what favourable finding would justify an action that is currently withheld.

The adverse finding deserves particular attention because it reveals whether the institution has made an evidence-contingent decision. If no possible result can change the commitment, the study is serving confirmation, communication or compliance purposes. Those purposes may still be legitimate, but their cost should not be justified by a claim of improved action selection. A charter should allow the author to state that the intended use is communication, provided the report does not then imply that a conditional decision rule was tested when none existed.

The favourable finding can be equally revealing. A positive result may be insufficient because the organisation still lacks financing, a permit, an operating capability or a contractual right. In that case the action menu should include the feasible next step, such as a bounded engineering investigation, rather than a full commitment. This preserves the difference between an attractive state of the world and an executable institutional response. The paper later shows how removing an unavailable action can materially reduce information value even when the signal itself is unchanged.

Finally, the charter should nominate a person who can challenge the loss function. An analyst can faithfully calculate a model that excludes the cost borne by another department or treats a protected interest as a compensable inconvenience. The person approving the research must know whose payoffs are represented, who bears omitted consequences and who may reject the proposed trade-off. Information value is conditional on that perimeter. A technically correct result can still be an institutionally unacceptable basis for action.