Open Sky Leadership Institute · Module 03
Safe-to-Fail
Experiments
In complex environments, the smallest probe teaches more than the largest plan.
Section 1 · Big Bets vs. Small Probes
Why large-scale change so often fails
Most organizations approach complex challenges the same way they approach complicated ones: with large, carefully planned interventions. A new strategy. A restructuring. A culture change initiative with a 36-month roadmap. Investment in the millions.
And most of the time, these large interventions produce exactly what complexity predicts they will: unpredictable results, unintended consequences, and organizational exhaustion.
The problem is not ambition. The problem is the mismatch between the type of challenge and the type of response. In complicated systems, large well-planned interventions work — because causes and effects are analyzable in advance. In complex systems, they don't — because the relationships between causes and effects only become visible after you act.
The question is not "What is the best plan?" The question is "What is the smallest action that will teach us something?"
Safe-to-fail experiments are the alternative. They are small, bounded actions designed not to prove a hypothesis but to generate learning. They do not assume you know the answer in advance. They assume the system will reveal itself — if you create the conditions to listen.
The goal is not to eliminate the risk of failure. The goal is to make failure fast, cheap, informative, and reversible — so that each probe moves the system forward, regardless of outcome.
Reflection prompt
A plan that didn't survive contact
Recall a significant organizational initiative that didn't produce the intended results. In retrospect, was the challenge complex or merely complicated? What would a smaller probe have revealed earlier?
Section 2 · The Right Kind of Wrong
Not all failures are equal
Amy Edmondson's research on organizational failure reveals something counterintuitive: the organizations that learn the most are not the ones that fail least — they are the ones that fail in the right ways. Her framework distinguishes three types of failure, each calling for a different response.
Basic Failure
A known process was not followed correctly. Cause and effect are clear. Prevention is the goal.
Example: A safety protocol was skipped and an injury resulted. This is preventable and should not happen.
Complex Failure
Multiple factors combined in unpredictable ways. No single cause, no single fix. Requires systemic analysis.
Example: A product launch underperformed due to market timing, team dynamics, and a competitor move — all at once.
Intelligent Failure
A thoughtful experiment in new territory that produced unexpected results. This is the right kind of wrong. It advances knowledge and should be celebrated.
Example: A pilot program surfaced a hidden assumption about customer behavior that redesigned the entire approach.
Safe-to-fail experiments are explicitly designed to produce intelligent failure. They are not failures of execution. They are deliberate probes into unknown territory, run at a scale where the learning outweighs the cost of being wrong.
Key concept
The goal of a safe-to-fail experiment is not success or failure — it is learning. A probe that confirms your hypothesis and a probe that contradicts it are equally valuable, as long as both were designed with enough intentionality to produce clear signals.
Section 3 · Designing an Experiment
Seven elements of a well-designed probe
A safe-to-fail experiment is not a vague intention to "try something." It is a structured action with seven specific elements — each of which forces clarity about what you are testing and how you will know what you've learned.
1. Name
A short, memorable title. The name should make the probe immediately recognizable in conversation and documentation.
2. Action Description
Who does what, with whom, for how long? Specific enough that someone else could run it without your presence. The more concrete, the more useful the learning.
3. Coherence Rationale
Why does this action make sense given what you know about the tension, the system, and the context? This prevents random experiments from masquerading as strategic probes.
4. Signs of Success
What would you observe early if this is working? Be specific — behaviors, conversations, data — not outcomes.
5. Signs of Failure
What would you observe early if this is not working? Name signals, not just absence of success signs.
6. Amplify
If you see success signals: how will you scale this up? What resources would you add? Who else would you bring in?
7. Dampen
If you see failure signals: how will you reverse course quickly and safely? What makes this safe-to-fail?
Reflection prompt
Where in your work is a probe more appropriate than a plan?
Name a challenge in your current context that has resisted traditional planning approaches. What would it look like to design a small probe instead?
Section 4 · Reading the Signals
The discipline of noticing
A safe-to-fail experiment produces learning only if someone is paying attention to what happens. This sounds obvious — but in practice, most organizations run probes and then get distracted before they've extracted the learning from them.
Reading signals well requires three disciplines:
- Separate observation from interpretation. Write down what you actually observed — behaviors, conversations, data — before you explain what it means. Collapsing these two steps is one of the most common sources of learning failure.
- Look for weak signals. The early signs of both success and failure are almost always subtle. If you're waiting for confirmation before acknowledging a signal, you've probably already missed the learning window.
- Make the signals social. Don't read them alone. The most valuable learning from a probe happens in a conversation with people who saw it differently than you did. Diverse observations produce richer accounts.
The discipline is not running the experiment. The discipline is staying curious long enough to hear what it says.
One practical tool: the signal check-in. Before the experiment ends, schedule a 30-minute conversation with the people closest to the probe. Ask each person: What did you observe? What does it make you think? What would you do differently? Write it down before you evaluate it. This conversation, done well, is often more valuable than the experiment itself.
Section 5 · Application
Build your experiment
Use the canvas below to design a real safe-to-fail experiment for a challenge you are currently navigating. Work through all seven elements before evaluating whether the experiment is worth running.
Name
Action description — what will you do?
Who does what, with whom, where, and for how long?
Coherence rationale — why does this make sense?
Why is this probe a reasonable response to the tension or challenge you're navigating?
Early signs of success
Early signs of failure
Amplification — if working
Dampening — if not working
Leadership Discourse Foundations
The scholars beneath this practice
Safe-to-Fail joins action research, learning beliefs, psychological safety, and complexity-informed experimentation.
Follow the intellectual trail: open any Scholar Profile above, or study the full chronology, four theoretical streams, and seven Open Sky lineages in Module 00: Leadership Discourse →
Section 6 · Key Takeaways
What to carry forward
- In complex environments, large well-planned interventions routinely fail because the relationship between cause and effect cannot be known in advance. Small probes are not a retreat from ambition — they are the appropriate tool.
- Not all failure is the same. Basic failures should be prevented. Complex failures should be analyzed. Intelligent failures — thoughtful probes in new territory — should be celebrated and learned from.
- A safe-to-fail experiment has seven elements: name, action description, coherence rationale, success signals, failure signals, amplification plan, and dampening plan. All seven matter.
- The learning from an experiment only happens if someone is deliberately paying attention. Schedule signal check-ins. Separate observation from interpretation. Make the sensemaking social.
- A probe that confirms your hypothesis and a probe that challenges it are equally valuable — as long as both were designed to produce clear signals.
Closing reflection
What will you run in the next 30 days?
Name the experiment from Section 5 and commit to a start date. What is the one thing you would need to do this week to make it happen?