data warehouse or not ? when is it okay to use OLAP without a data warehouse database

Hello,

In the process of designing a cube, I would like to ask "when is creating a cube without a data warehouse database considered okay".
Does it depend on the database size ? on how frequent errors happen in the relational db ? on how frequently new tables are created in the relational db ?
(for info, in my case, the management expects to request a new BI report with recent data every month.)

Thank you.

The general consensus among all DW experts is you don't do that. It leads to a mess. Basically what happens is the first solution is built, the business likes it and it leads to other such attempts, with each new one doing its own extract. Eventually, the number of extracts gets out of hand, impacting performance of the operational system and becoming unmanageable from a coding and support standpoint.

But, if this is a pilot to test the waters, go ahead. Just go in with a clear understanding that this is basically throw away work and if the business want's to pursue a stable, productive BI environment on a larger scale, they need to do it right. Then again, usually that understanding is forgotten once the first mart is a success, and there is pressure to repeat it... quickly.

I totally agree with the previous comment.

Further to that, in my experience it is very difficult to meet the typical design, data quality and data enhancement requirements of a useful cube without the platform of a database containing the relevant dimensional model.

If you proceed without a dimensional model database you will probably burn a lot of time "trying to get the cube to produce the right results", which typically means trying to resolve problems with referential integrity, incomplete data etc. These are all problems best solved by a database in a dimensional model.

Good luck!
Mike

Thank you both for your replies.
I have recently joined a company where different development departments each create their own cubes independently (without DW).
My multi-tier web development background convinces me enough that a DW (intermediate layer) is definitely beneficial. However the management will not be convinced until a quantifiable measure of the improvements over the costs is available.
As you said, I will work with what I currently have (actually trying to fix a DW-less cube by comparing its reports to other reports generated from the relational DB), but also will work on quantifying the benefits of including a DW in the architecture.