When you’re pricing/packaging an AI feature or approaching launch and need to ensure margins won’t get destroyed by inference/tooling costs.