I‘d probabl argue that not all of those 90% must be considered non-fancy. Deploying models and telling business folks about ML for example. ;)