Flip a fair coin a million times. You have a million data points, the chances of getting any particular sequence is astronomically small), there are a vast number of combinations whether order matters or not, etc. Yet every sequence has exactly the same chance, ever data point has only two possibilities/dimensions, ever toss is clearly either heads or tails, everything is simple.

Try running component analyses or other compression algorithms on a REALLY big image file and see how many hundreds of thousands of data points you can remove before you can notice the difference.

The "law of large numbers" has (usually improperly) used to defend the use of a sample of 30 people to represent millions. Actually have lots of numbers can simplify things enormously, not complicate them (thanks to computers). Relatedly, networks governed by "small world" dynamics can washout or not exist globally but exhibit complex, unpredictable, and require models with lots of variables. The internet is becoming increasingly hidden (the "dark web" or "invisible web" phenomenon) to google searches because so many sites are like this one: dynamically generated & continuously changing with many links appearing and disappearing hour by hour. Small world dynamics are making vast data acquired by web-crawling using the most sophisticated algorithms and technology around end up missing enormous amounts of what's out there.

The key, given any business plan, any market research, any compiling of some database or use of cloud computing or whatever, is knowing what kind of data is actually important an its nature. You can greatly oversimplify just about anything almost as easily as you can needlessly overcomplicate things. The first book (published by CRC press) on statistics and "big data" database marketing I bought out of curiosity. I know the publishing company, and they aren't prone to publishing sensationalist bunk, so I was surprised to find a book with a subtitle including the phrase "mining big data" to be nothing more than an intro stats book with some novel terms, some less-than-usual methods included, and a lot omitted that should have been there. Yes, we live in a world of vast amounts of information flowing all the time. Most of it is meaningless and useless for most purposes.

Do those businesses have to handle PB of data each day? Are those sets of data changing at a speed rate that their systems can't cope with? Does the former imply a loss of money?

If so they could get some advantage of the process of adapting their hardware and software and technicians.

If not, for sure they can continue with their hardware and software tools for Data Mining.

I think Andrew has made an excellent point. Big Data is everywhere around us all the time now. Something like 90% of all the digital data that exists has been generated in the last decade, and this rate of data production will increase dramatically. However, data is not the same as information. I would define information as having some practical context and purpose. Unless you can articulate that context/purpose, then it doesn't matter how much data you have, you will probably never obtain value from it.

In my opinion, this is where people may become "caught-up" in the thought that they MUST have Big Data... and yet they won't necessarily know what to do with it once they have it.

Big-data may have its pros - but from a customer perspective, big-data is a 'disruptive' technology, at least that's how the they see it (or made to see it).

We regularly deal with customers who are on both sides of the river - those who are very much interested in bigdata (without having a clue of what it is all about), and then there are those who are not at all interested in making a move towards bigdata (despite the fact that they know the benefits of it).

The main problem comes from the disconnect between the marketing teams and their understanding of the technology. Also, many a customer considers it more as a marketing hype, thanks to the lack of substantial concrete evidence the marketing teams have when they try to sell on how big-data really helps the business.

This is not to say big-data does not have advantages - only that the customers do not see it yet. For example, here is another question https://www.researchgate.net/post/References_for_Big_Data_Success_Stories that talks about the lack of "availability" of success stories in general on bigdata. My answer to that question has some useful info.

A very interesting question. I would like to answer from a different perspective: In my profession, I have always dealt with SQL system and I must say that I can do my job well with it, never thought of a different approach, at least until now. Reasons behind:

a. I am not working with EXTREMELY large data sets.

b. Most of my data is structured.

I believe that one should not think of, let´s say, SQL “OR” Hadoop/MapReduce/Spark, rather one should think of SQL “AND” Hadoop/MapReduce/Spark.

Traditional relational databases will not disappear. What is happening is that nowadays we deal with a variety of data, and this implies both structured (intended for computer processing) and unstructured data.

What represents a better fit, is context-dependent.