Value of flexibility and decision timing

An investment choice need not be “build everything now” or “reject forever.” A company may buy information, build a smaller stage or retain a right to expand after demand becomes clearer. Flexibility is valuable when new information changes an available decision and the benefit exceeds the cost of waiting and testing. Calling a plan “flexible” is insufficient. Draw a decision tree with the date of each payment, possible states, conditional cash flows and actions available after a signal. If contracts or reserved capacity make stopping or expanding impossible, do not assign value to an option the business does not own.

Expected value is the probability-weighted average of possible outcomes. It helps comparison under suitable assumptions, but it neither guarantees an individual outcome nor captures the firm's ability to survive a loss. Start at the last decision node: after observing the signal, select the higher-value available action in that state. Work backwards, weighting state outcomes by their probabilities. Do not use tomorrow's information in a decision taken today. When branch values are net present values measured at a date, make their dates consistent. Adding a value at the end of next year to today's cost without discounting is a common error.

A fictional company is considering a plant costing 60 now. If demand is high, the present value of its cash inflows is 105; if demand is low, it is 45. Management's illustrative probabilities are 55% high and 45% low. NPV is 105−60 = 45 in the high-demand state and 45−60 = −15 in the low-demand state. Expected NPV of immediate investment is 0.55×45+0.45×(−15) = 18. Positive expected NPV conceals a chance of losing 15; compare that downside with liquidity and borrowing capacity. No investment is worth zero in this simplified model, although a real lost customer or rival opportunity could make inaction costly.

Operations proposes a one-year test with a present cost of 4. A “strong” or “weak” signal appears at the end of the year, when management may build or refrain. To show the method, suppose the present value today of the net cash flows from building after the test, with all next-year payments discounted back, is 34 after a strong signal and −8 after a weak one. The chance of a strong signal is 50%. Those are assumed outputs of delayed production, demand learning and payment dates; they are not 45 and −15 with new labels. Finance would need to show the schedule behind them for a real approval. Build after a strong signal because 34>0. Refrain after a weak signal because zero beats −8. The test plan is worth −4+0.5×34+0.5×0 = 13 today. Testing may be preferable to later uninformed investment in some circumstances, but here it trails the immediate value of 18. The gap of 5 is the unrecouped opportunity cost of waiting under these assumptions. Do not praise a test for “reducing risk” while ignoring what delay costs.

The conclusion may change if the firm cannot absorb the low-demand loss. Suppose a liquidity policy prevents an option that could cause a first-year cash loss above 10, and −15 represents an unfinanceable cash outflow, whereas the test limits the certain loss to 4 and allows exit after a weak signal. Immediate investment then fails the liquidity gate despite its higher expected NPV. Testing at value 13 beats not investing if its own downside is acceptable. Verify that −15 truly describes cash at the stressed date rather than only NPV. A project can have negative NPV yet enough first-year cash, or positive NPV with an early cash deficit. A real decision requires dated cash obligations and funding scenarios. The example illustrates the role of a binding constraint rather than a universal policy.

What is information actually worth?

Separate signal accuracy from delay when judging an improved test. If a free, perfectly accurate signal arrived now without delay, management would build only in high demand. Expected value would be 0.55×45+0.45×0 = 24.75, which is 6.75 above the immediate build decision of 18. In this model 6.75 is the upper bound on the gross value of perfect information if nothing else changes. A test costing more cannot be justified by information alone. The actual test above is neither perfect nor free nor immediate. Comparing its 13 directly with 24.75 does not prove that accuracy alone destroyed 11.75; delay and different post-test cash-flow estimates also matter. Separate those effects to decide whether to improve the test, shorten it or abandon it.

The 55% high-demand probability is an analytical assumption, not a certificate from the market. Test it. Immediate investment value as a function of high-demand probability p is p×45+(1−p)×(−15) = 60p−15. It ties no investment at p=0.25. It ties the test value of 13 at p=28÷60≈46.67% if the test's values remain fixed, a strong assumption because signal probabilities may also change as p changes. If plausible demand probabilities range from 40% to 60%, the comparison moves: at 40% immediate value is 9, below the test's 13, while at 60% it is 21, above it. This is a sensitivity analysis, not a statistical verdict about p. Collect customer reservations, cancellation conditions and rival capacity; update the tree rather than preserving a single probability forever.

Do not add an arbitrary discount-rate penalty to every branch after already representing adverse outcomes in state cash flows and probabilities; that can double count risk. Nor should expected value alone replace a survival decision. A board may decline a positive expected-value project that threatens the firm's continuation, but it should state the survival constraint and compare the cost of alternatives. Flexibility does not excuse vague escalation. Set a rule for moving from the test to a plant: confirmed orders, contribution margin after discounts, a working-capital cap and an expiration date for the expansion right. If building costs rise during the wait, update the branch value rather than merely rewording the memo.

Solution: immediate expected NPV is 0.4×(55−30)+0.6×(22−30) = 10−4.8 = 5.2. No investment yields zero. With a test costing 3, build only in the strong market: −3+0.4×25+0.6×0 = 7, which exceeds immediate investment by 1.8. Gross perfect-information value is 10−5.2 = 4.8; after its 3 cost, the improvement is 1.8. If testing costs 8, its plan value is −8+10 = 2, below immediate investment at 5.2. Real signals are usually imperfect and take time; they may change supplier prices or competitors' behavior. Then model conditional signal probabilities, update beliefs after the signal and date every cash flow. Never insert a strong result observed after a test into a decision node before testing, or describe the 5.2 expected value as guaranteed cash.

Additional decision lab: the price of conditional waiting

Suppose the board of a fictional manufacturer can build a new line now for 60. The present value of future operating cash flows is 105 if demand is high and 45 if demand is low. Initially each state has a 50% probability. Expected immediate NPV is 0.5×105+0.5×45−60=15. The average hides an important asymmetry: with advance knowledge of low demand, the company would refuse to build. Management can run a market test for 4 and delay the construction choice by one year. The test produces a strong signal with 50% probability. Conditional probability of high demand is 80% after a strong signal and 20% after a weak one. These fictional probabilities are coherent because they imply an unconditional high-demand probability of 50%. For the comparison, assume the stated cash-flow values are measured at the investment date one year ahead and the 60 is paid on that date. Discount that future decision value to today at an illustrative 10%; do not compare values at different dates.

After a strong signal, expected operating value at the decision date is 0.8×105+0.2×45=93, giving net construction value 33. After a weak signal it is 0.2×105+0.8×45=57, so net construction value is −3 and the firm uses its right to decline. Expected optimized decision value next year is therefore 0.5×33+0.5×0=16.5. Its value today net of test cost is 16.5÷1.10−4=11. This test does not beat immediate construction's 15 under the stated, date-consistent assumptions. “Information has value” does not mean buying any information is worthwhile. Even with an instantaneous test, 16.5−4=12.5 remains below 15. The board should challenge the test's cost, delay or accuracy, or compare a cheaper contractual option.

The conclusion could change if stopping after construction carries an unmodeled penalty or if the test allows a smaller investment in weak demand. Do not add a generic “option value” to NPV while separately counting abandonment in the scenario tree. That would count the same flexibility twice. Draw the decision tree with the date of information, the action then available, and the cash already irrecoverable. A common probability error is to confuse the probability of receiving a strong signal with the probability that demand is high after receiving it. They are not the same statement. Another error is to value the test against an artificially weak alternative. Compare it with immediate building, waiting without testing, and walking away; the best alternative sets the hurdle.

An effective board memo would include a reversal threshold: the maximum test cost or delay consistent with choosing the test. It would identify the data needed to produce the signal, the owner of the next decision, and the deadline for exercising the build right. A supplier contract that allows cancellation after a small deposit may buy flexibility more cheaply than a separate research project. Conversely, if the board would build regardless of test outcome, the test has no decision value even if it generates interesting information. The question is whether information changes an available, consequential choice, not whether the team has labeled its plan “staged.”