I think it's also a bit scary, because 8 weeks is very little time for testing, tuning, and validation of something as opaque as a machine learning model. If it worked right the first time, that's great. But there is still a lot of inherent uncertainty in ML projects. Decision makers need to take that uncertainty into account when planning.

That, or, the 8 weeks only covers the final training runs and the implementation/deployment, and doesn't include time spent developing and tuning proof-of-concept prototype models.