People make work possible in conditions that rarely match a procedure perfectly. They coordinate, notice weak signals, adjust the sequence, and work around friction. A Learning Team creates room to understand those conditions with the people closest to them. There does not have to be an incident first.
Choose a useful question
Start with a task, a repeated difficulty, a change in equipment or staffing, or a job that regularly depends on extra effort. Frame the invitation around curiosity: “How does this work happen on a demanding day?” is more useful than “Why don’t people follow the procedure?”
Bring together people who perform and support the work. Ask them to describe a recent example in concrete detail: the goal, what they could see, the tools and time available, where the plan fit, and where they had to adapt. A facilitator should make space for different accounts and avoid turning the first conversation into a search for fault or a rush to solutions.
Move from understanding to action
A useful pattern is to separate the learning conversation from the action conversation. Between them, participants can check details, observe the task, and think about what was missed. The second conversation can then test possible changes against the actual conditions of the job.
- Explore: What helps the work succeed? What makes it harder?
- Reflect: Which assumptions should be checked with evidence or observation?
- Design: What change would make the work easier to perform safely?
- Verify: Who will see how the change works in practice, and when?
The number of meetings is less important than preserving that sequence. The team should know what it learned, what remains uncertain, who owns the next action, and how its members will hear back. Without follow-through, asking people to share what they know can erode trust.
Keep the methods distinct
After an event, an investigation may need to establish a supported account and test controls. A Learning Team can then explore a wider pattern of everyday work when the account raises questions it cannot answer. It can also start independently, before harm occurs, with a question about how a task succeeds.
Adapted and edited from Todd’s 2024 article archive.