moheetsubudhi-isb/decision-analysis-toolkit

Decide between options whose payoff depends on uncertain outcomes, using a decision tree and expected monetary value (EMV). Always use this skill when someone asks whether to launch, invest, expand, bid, settle, build or drill and gives, or can estimate, payoffs and the chances of each outcome, even when the question sounds like a simple yes or no. Also use it to lay out decision and chance nodes and fold the tree back; compare several risky options; test how sensitive the choice is to a probability or payoff, or find the breakeven probability where it flips; or decide whether risk aversion, a utility function or a worst-case view should override expected value. Not for decision tree, random forest or boosting models trained on data, not for pricing a test, survey or pilot before deciding, not for simulating many uncertain inputs, and not for explaining what a decision tree is when no decision is in play.

Design, run and read a Monte Carlo simulation when a plan depends on several uncertain inputs. Use whenever someone asks for a risk or what-if model with uncertain demand, cost, price, duration or returns; the chance a project, budget or forecast misses its target; a range instead of a single estimate; which distribution to use for an input; how many simulation runs are enough; how much to order or stock for one selling season when demand is uncertain, including the newsvendor critical ratio from overage and underage costs, and the service level or fill rate it implies; the probability a project finishes by a date or which tasks drive schedule risk; or why a plan built on average inputs is too optimistic. Not for choosing between a few options with known probabilities in a decision tree, not for deterministic optimisation of a production or allocation plan, and not for forecasting a time series.

Decide whether to buy information before a decision: a market test, pilot, survey, trial run, inspection, diagnostic, consultant report or extra data. Use whenever someone asks whether a test or study is worth its cost; the most they should pay for research, a pilot or better data; the expected value of perfect information (EVPI) or of sample information (EVSI); how an imperfect test with false positives and false negatives should update the odds; why a positive result from an accurate test can still mean the event is unlikely (base rates, Bayes' rule); or whether to decide now or wait and learn. Not for building the decision tree for a choice with no test on offer, not for A/B test sample size or significance, and not for designing the survey questionnaire itself.