If you’re building experimental GenAI features that haven’t proven their product market fit, you don’t want to commit to a model that runs up costs without a return on that investment.