When to Stop Researching: The Expected Value of Information

Jagdeep Singh · Jagdeep Ventures · Last revised July 2026

Every research effort has a silent price tag: the cost of the work, and the cost of the delay. Decision science offers a discipline most operators never meet, the expected value of information, that turns "should we study this more?" from a comfort question into an arithmetic one.

The core idea

Information has value only when it can change your decision. If you would enter the market whether the study says the segment is worth $80M or $200M, the study, whatever its intellectual merits, is worth nothing to this decision. The expected value of perfect information (EVPI) formalizes this: it is the difference between the expected outcome of deciding with perfect knowledge and the expected outcome of deciding now with what you have. It is the absolute ceiling on what any research can be worth, and it is frequently, usefully, embarrassingly small.

Real research is imperfect, so its value is lower still, the expected value of sample information (EVSI). You rarely need to compute either precisely. The discipline is in the framing: before commissioning analysis, write down what you would do under each plausible finding. If the action column is constant, stop.

A worked example

Suppose you are deciding whether to launch a product variant. Launching costs $250k; you estimate a 60% chance it contributes $600k and a 40% chance it contributes roughly nothing. Expected value of launching: 0.6 × 600 − 250 = $110k, versus $0 for not launching. Now: what is a perfect market test worth? With perfect information you would launch only in the good state: 0.6 × (600 − 250) = $210k. EVPI = 210 − 110 = $100k. A $40k study that resolves most of the uncertainty is a bargain; a $150k study is a donation to the research industry, and a three-month delay may cost more than either.

The playbook

1. State the decision and the alternatives first. Not the question, the decision. "Understand the market" is not a decision; "enter, wait, or pass" is.
2. Write the action table. For each plausible research finding, what would you actually do? Findings that map to identical actions are decoration.
3. Estimate the stakes spread. How much worse is the worst plausible action than the best, in expectation? That spread bounds what any information is worth.
4. Price the delay. Research costs money and time; in competitive situations the time is usually the expensive part. Subtract it.
5. Buy the cheapest decisive test. Rank open questions by how much a resolution would move the decision per dollar and week. Fund the top one or two. Ignore the rest until the decision changes.
6. Set a stopping rule in advance. "We decide on the 15th with whatever we have" outperforms open-ended inquiry for most business decisions, because the marginal value of information decays fast while its cost does not.

Where this bites in practice

Three patterns recur in our engagements. First, confidence laundering: research commissioned not to inform a decision but to arm a decision already made. EVPI is zero by construction; the honest move is to say so and spend the budget on execution. Second, symmetric anxiety: teams research the downside exhaustively while leaving the upside unexamined, though the value of information is highest where the decision is genuinely poised. Third, precision theater: buying a tighter confidence interval on a variable the decision is insensitive to. A tornado chart, sensitivity of the outcome to each assumption, is the antidote, and it costs an afternoon.

The uncomfortable conclusion of the framework is that the right amount of analysis is usually less than analytical people want and more than impatient people tolerate. The framework does not resolve that tension; it prices it, which is what a decision discipline is for.