With the use of a machine learning algorithm, predictive scaling is getting more accurate, but you have to deal with sudden traffic spikes, and depend upon reactive scaling. This unexpected traffic that may come could be even 10 times the regular traffic; this usually happens due to a sudden demand or, for example, due to a first attempt to run sales events, where we're not sure about the amount of incoming traffic.
Let's take an example where you are launching a flash deal on your e-commerce website. You will have a large amount of traffic on your home page, and from there, the user will go to the flash deal product-specific page. Some users may want to buy the product; therefore, they will go to the add to cart page.
In this scenario, each page will have a different traffic pattern, and you will need to understand your existing architecture and traffic patterns, along with an estimate of the desired traffic. You also need to understand the navigation path...