# Failing to Learn and Learning to Fail (Intelligently): How Great Organizations Put Failure to Work to Innovate and Improve

*Cannon · Safety & Error, Team Learning · Long Range Planning · 2005 · Open access*

A practitioner-facing synthesis of the authors' prior field research, predating and considerably more detailed than Edmondson's later and more famous 2011 HBR piece on the same territory ('Strategies for Learning from Failure', which distills much of this article's core argument for a wider audience), building a framework around two types of barrier and three organisational processes. Defines failure broadly as any deviation from expected results, spanning technical mistakes to interpersonal ones, and includes both avoidable errors and the unavoidable negative outcomes of legitimate risk-taking. Technical barriers to learning from failure include a lack of the basic analytical skill to draw valid inferences from experience and task designs that hide failure, such as excess work-in-process inventory in manufacturing, which delays discovery of defects until many units have already been produced the same flawed way. Social barriers run deeper: people have a documented tendency toward 'positive illusions,' unrealistically favourable self-views that are otherwise a marker of good psychological health but actively work against honest acknowledgement of failure, and Finkelstein's research on major corporate failures found that more senior executives tend to externalise blame more, not less, than junior staff. The paper's most useful diagnostic distinction is between small and large failures: organisations reliably over-attend to catastrophic failures via after-the-fact investigative commissions while chronically under-attending to the small, everyday ones that are usually the 'early warning signs' that would have prevented the catastrophe, illustrated by Mattel's Jill Barad missing earnings guidance for four consecutive quarters without acknowledgement and the collapse of Australia's HIH Insurance Group, where management actively concealed early problems from its own board. The framework sets out three learning processes, presented in order of increasing organisational difficulty but explicitly not meant as a strict sequence: identifying failure, analysing failure, and deliberate experimentation. Identifying failure is illustrated by Dr. Kim Adcock at Kaiser Permanente, who used longitudinal mammogram-reading data to convert an otherwise invisible baseline error rate (10-15% even among expert readers) into individualised, actionable feedback for radiologists, contrasted with NASA's 16-day failure to identify the Columbia foam strike as a genuine problem at all. Analysing failure is illustrated by Julie Morath's Patient Safety Steering Committee at Minneapolis Children's Hospital, which extended formal 'Focused Event Studies' down to small near-misses rather than only major accidents, and by pharmaceutical R&D's habit of repurposing failed drugs (Pfizer's Viagra was originally an angina treatment; Eli Lilly's Evista began as a failed contraceptive and Strattera as a failed antidepressant), most strikingly in the case of Eli Lilly's Alimta, an experimental chemotherapy drug rescued from an apparently failed trial when a mathematician whose actual job was investigating failures discovered that adverse reactions correlated with folic acid deficiency, solved simply by co-administering it. The night-before teleconference preceding the Challenger launch decision is the central illustration of failed analysis: Argyris's research on why people in disagreement rarely ask each other genuinely sincere questions plays out directly in the transcript, where engineers offered abstract, unpersuasive assertions ('it is away from goodness to make any other recommendation') instead of walking administrators through the underlying data, and the discussion polarised rather than converged. Deliberate experimentation is framed as the 'offensive' counterpart to the first two, more defensive processes: deliberately manufacturing more failure as the necessary cost of discovering genuine novelty, illustrated by IDEO's 'fail often to succeed sooner' culture, PSS/World Medical's 'soft landing' policy protecting employees who try an internal role change and don't succeed, and 3M's explicit target that 25% of divisional revenue come from products launched in the previous five years. Bank of America's real-branch innovation 'laboratories,' with an explicit target failure rate of 30% set as a positive signal of genuine experimentation, doubles as a cautionary tale: employees remained reluctant to experiment until management fixed the mismatch between this stated goal and reward systems that still primarily measured routine performance. The resulting framework crosses the three processes against the two barrier types to yield six concrete recommendations, from building anomaly-detection information systems and structured after-action-review formats on the technical side, to blameless reporting systems, skilled facilitation for failure-analysis discussions, and reward systems explicitly realigned with a stated tolerance for experimental failure on the social side. Closes with organisational scholar Sim Sitkin's five criteria for a failure actually being 'intelligent' rather than merely careless: it results from thoughtfully planned action, has a genuinely uncertain outcome, is modest in scale, is responded to with alacrity, and occurs in a domain familiar enough to permit real learning from it, alongside a broader table reframing the traditional managerial mindset (failure as unacceptable, self-protective responses, cost control) against a learning-oriented one (failure as the natural byproduct of experimentation, curiosity, investment in future capacity).

- **This page:** https://explore.psychsafety.com/n/cannon-edmondson-2005/
- **View the source paper:** https://blog.educpros.fr/francois-fourcade/files/2014/06/Article.pdf
- **Interactive map:** https://explore.psychsafety.com/?mode=papers&node=cannon-edmondson-2005

## Connected concepts (12)

- [Normal Accidents: Living with High-Risk Technologies](https://explore.psychsafety.com/n/perrow-1984.md) (paper)
- [Disrupted Routines: Team Learning and New Technology Implementation in Hospitals](https://explore.psychsafety.com/n/edmondson-bohmer-pisano-2001.md) (paper)
- [High-Quality Relationships, Psychological Safety, and Learning from Failures in Work Organizations](https://explore.psychsafety.com/n/carmeli-gittell-2009.md) (paper)
- [Managing the Risk of Learning: Psychological Safety in Work Teams](https://explore.psychsafety.com/n/edmondson-2002.md) (paper)
- [Speaking Up in the Operating Room: How Team Leaders Promote Learning in Interdisciplinary Action Teams](https://explore.psychsafety.com/n/edmondson-2003b.md) (paper)
- [Why Hospitals Don't Learn from Failures: Organizational and Dynamics That Inhibit System Change](https://explore.psychsafety.com/n/tucker-edmondson-2003.md) (paper)
- [Blametropism](https://explore.psychsafety.com/n/blametropism.md)
- [Errors in Organizations: An Integrative Review via Level of Analysis, Temporal Dynamism, and Priority Lenses](https://explore.psychsafety.com/n/lei-naveh-novikov-2016.md) (paper)
- [Learning from Error](https://explore.psychsafety.com/n/learning-from-error-or-punishing-it.md)
- [Learning Teams](https://explore.psychsafety.com/n/learning-teams.md)
- [Linking Leader Inclusiveness to Work Unit Performance: The Importance of Psychological Safety and Learning from Failures](https://explore.psychsafety.com/n/hirak-et-al-2012.md) (paper)
- [Watermelon Effect](https://explore.psychsafety.com/n/the-watermelon-effect-and-greenwashing.md)