A large part portion of the software industry skips this form of testing in favor of using a portion of production with health metrics to detect chaos. Which is alright, unless someone like Boeing does it with an airplane.

If unleashing chaos on users is bad enough for your business, then TLYF is a good pattern. Here are some scenarios that call for TLYF:

* Customer will die from chaos in software

* Customer will remove product if they experience any chaos

* Software is deployed somewhere that is hard to update - think non-internet connected devices, or regulatory required change control

* Chaos will cause significant loss of value

Both Boeing and Toyota failed to apply TLYF rigorously, resulting in uneccessary deaths.