New Agora

Complex environments are best described as a system where cause and effect are not obvious, and may not ever be fully known. In these environments, traditional tools such as outcome-based project management and scientific methods are a poor fit.

Safe-fail experimentation is a problem solving technique that emphasizes controlled failure through the conduct of many varying experiments. It is particularly effective in a complex environment as a method of learning.

With acknowledgement to Raymond Salzwedel, Dave Snowden and Joe Firestone for much of the thinking and debate, here are some of the basic principles of safe-fail experimentation:

1. Experiment freely and expect failure. Failure is good and promotes learning.
2. Try different experiments in the same environment and the same experiment in different environments. In complex systems, the changed context of a different environment can lead to very different outcomes.
3. Start with experiments where failure can be tolerated. That is, choose experiments where the overall impact of failure on the system is likely to be small and/or manageable.
4. Design experiments that can be monitored. To plan future experiments, you need be able to determine whether the outcome of an experiment was favourable.
5. Run multiple experiments in parallel. In safe-fail mode, this is not only permissible, but encouraged where experiments can be isolated and results independently evaluated. Over time, experiments producing undesirable results should be wound up and new experiments started in promising areas.
6. Share the results of your experiments with others, and learn from the results of their experiments. This is just a logical extension of the idea to run experiments in parallel. You must be cautious, however, because the differences between your context and theirs may be difficult to isolate. Therefore, do not assume that someone else's results can be replicated within your system - but do