AFAIK they specialize in making data aggregations (for relational data mostly). That is data that is preprocessed in some way to reduce its volume. An example can be a log file with timestamps accurate to a second. You rarely need that acuracy - so you can sum the number of records of each type in every hour and keep just the information of those sums. You'll still be able to extract any information you need from this data, but the you reduced the volume of the data - and those extractions would be much faster.