Which failures are worth running on purpose

The portfolio review hears what landed. A small test that missed its guess was taken off the page last week. A skipped check on a live process was taken off too. The room has no way to tell those two events apart.

Amy Edmondson sets the distinction out in a December 2023 note for Harvard Business School Executive Education, drawing on her book Right Kind of Wrong. She defines failure as an outcome that deviates from desired results. Mistakes are avoidable errors. Failed experiments belong to new territory. She describes two cultures. One talks as if failure is not an option, and she says that culture produces fear, less risk-taking and hidden failures. The other repeats fail early and often, or fail fast and break things. She says experiments are not useful in every situation, and that undisciplined experimentation produces activity with little benefit. Neither approach, on its own, leads to long-term success. The work is to match the approach to the challenge.

The three types

Her findings sort failure into basic, intelligent and complex.

A basic failure sits on a process the organisation already understands. The task is to follow the steps, and the mistake could have been caught, corrected or prevented. People stray from a known process, rebuild a step because they never found the existing practice, skip a step or break a practice on purpose. Prevention, in her account, is clearer communication and training that makes the known step easier to follow. A hospital that scans a wristband and checks the patient's name and date of birth, so the medication matches the order, is her example.

An intelligent failure has no settled process to copy. A new business model and a new kind of drug are the cases she uses. You test a provisional approach, some tests miss, and the next test is designed from the lesson. Before a miss counts, she asks whether this is new territory, whether the opportunity is worth the risk, whether you stated a hypothesis from prior knowledge, whether the experiment is as small as it can be and still informative, and whether you shared the lesson.

A complex failure has more than one cause. Oil spills and collapsed buildings are the scale she names. People want a single cause and someone to blame. She calls that instinct unhelpful, because small factors line up, and she says these failures are becoming more common as businesses depend on complex information technology. Study the ones you have had. Thank the person who raises an early warning, knowing some warnings will be false alarms. She says voice has to be cultivated, or a crucial warning is lost.

Where an AI miss sits

Label the miss before anyone proposes a consequence. A known check that nobody ran on a model output is a basic failure, so put the check on the task. A small test of a new use, with a stated guess and a written lesson, can be an intelligent failure when it misses. A customer-facing miss that took a bad prompt, a missing owner and a quiet reviewer is a complex failure, and the other factors stay if you stop at the first name you can find.

Edmondson calls psychological safety the antidote to the fear of being wrong. People can then run a thoughtful experiment and say when they see a mistake, before it grows. A healthy failure culture rewards an intelligent failure. If you only hear good news from the people who report to you, they may feel unable to bring a problem, a concern or a question.

1. Label each miss basic, intelligent or complex before anyone proposes a consequence.
2. For a basic miss, name the known step that was skipped and put that step on the task.
3. For an intelligent miss, require the hypothesis, the small size of the test and the shared lesson.
4. For a complex miss, list the factors before you name a culprit, and thank the early warning.

A psychological safety workshop can give an intact team a place to practise that sorting on a real miss. The next review has to hear a skipped check and a small experiment as different events.