What if we built a system that could automatically scale up or down depending on what kind of traffic we *expect* to

we're all familiar with auto-scaling based on current metrics like CPU or memory. But what if we could predict future load? Imagine a system that could proactively scale up *before* a known spike (like a marketing campaign launch or Black Friday) or scale down *after* a predicted lull. What kind of data would we need to feed into such a prediction model? What are the challenges in building and maintaining this, and are there any existing services or patterns that tackle this 'predictive scaling'?