Can Predictive modelling be applied to only one problem or multiple problems

I came across a company who does predcitive modelling and they say

开发者_开发问答

Because our Predictive Models are tuned to your unique business requirements and circumstances, they can be applied to a wide variety of forecasting and prediction activities across numerous problem types and domains

I want to know whether Predicive models are uniquely programmed for each task or they are one time programmed and used everywhere. I am bit confused

It varies: you could have a predictive model that is highly optimized to perform really well with a specific data set by taking advantage of things that are specifically know for the data. For example, if you're doing analysis of URLs and each URL is a string (with up to 255 characters), each (UTF 8) character is 8 bits or 1 byte so you would need 255 bytes to store each URL. You could hash each URL with a hashing function like CityHash64 and if you would now be able to store the URL with 8 bytes instead of 255 bytes. This could be built into the algorithm in order to make it use less memory which would allow it to fit more data into RAM and it would make the calculations faster, but this type of optimization cannot be made for a system that is intended to support various data types. In a general case, you would not have enough domain knowledge to make such optimizations.

On the "opposite" end, where this company seems to be, you would provide a generic system that takes in data in a certain format and applies predefined algorithm on the data. You may have some parameters that allow you to tweak the system in case you get some more domain knowledge, but it will be nowhere near like taking the hash of a URL in order to reduce the memory usage by 3000%. Companies that offer these kind of services usually apply the DRY principle, they code once and they reuse the code as much as they can (to save money and maximize profit). They may do some minor optimizations or tweaks in their settings, but they will generally avoid coding the algorithm just for a specific client.

This is by no means an accurate representation of what the company you're looking at is doing, but based on your quote it seems like a valid assumption. Just so people understand that I'm not passing judgment: it makes perfect sense if that company is generalizing, if my business was dependent on me servicing as many customers as possible then I would generalize too.