Information technology project failures: Applying the bandit problem to evaluate managerial decision making

This paper seeks to apply results from the study of bandit processes to cases of information technology (IT) project failures.

This paper examines three published case studies, and discusses whether managerial actions are in accordance with the predictions of bandit process studies.

Bandits are a class of decision‐making problems that involve choosing one action from a set. In terms of project management, the firm selects from several alternative IT projects, each with its own distribution of risks and rewards. The firm investigates technologies one by one, and keeps only the best‐performing technology. The bandit perspective implies that managers choosing a risky IT project with high potential reward before safer ones are behaving optimally. It is in the firm's interest to resolve the uncertainty about the innovative project first. In case of failure, the firm can later choose safer technology. A high proportion of risky projects adopted leads to a high number of project failures.

The bandit approach supports studies that advocate evaluating decision makers on the optimality of their decision process, rather than specific outcomes.

This paper demonstrates how insights from the bandit problem are relevant to studies of IT project failures. Whilst choosing high‐risk, high‐reward projects may be in a firm's interest, some observed project failures are optimal choices that do not work out.