Hi Chen,

I know of no other way to have people lose faith in a delivered DW faster than having the DW/DMs present differences of opinion about what a number should be.

Time and time again I have been asked to discuss with companies what to do with their array of 'independent, semi-dependent, interdependent data marts' because the answers coming from each of them were slightly different.

In nearly every case, senior managers refused to use the information at all if there was even the slightest hint that it was not consistent.

This is strange because they will use highly inaccurate information just as long as every data mart gives the same wrong information.

My answer, over years of experience came down to, if you have 4 or more, throw then away and start again. I was thrown out of plenty of places for proposing that course of action by the way.

The timing differences that will occur between operational systems and the data marts you are proposing will be more than enough to cause differences that may well undermine your efforts.....I'd always suggest that 'drill through' be 'drill through' to a detailed level that is synchronised with the data marts.....whether it be a so called EDW or just more detailed dimensional data. By all means drill through to operational systems if need be, but provide some kind of 'disconnect'

between the data marts and the operational systems so that differences in the numbers are far less likely to be brought up as a concern....

Best Regards

Peter Nolan

Data Warehousing Consultant

Mobile: +353 879 581 732

-----Original Message-----

From: owner-dwlist@datawarehousing.com

Sent: 27 July 2004 15:56

To: dwlist@datawarehousing.com

Subject: dwlist: Data inconsistent issues

From: "Chen, May" <May.Chen@FMR.COM>

Some of our designers/architects have the concept that detailed information must come from an operational database (ER model, a replicated transaction database) while analytics will be using data marts (Dimensional model). During an analytics session, we drill down to some level of detail and then the user will see a more detailed page that is sourced from operational database.

Here are some common sense issues:

The data from the 2 databases are on different frequencies - one is real time and the other is daily snapshot. The users may see inconsistent data. For example, on one report, I have 2 orders and next report shows that I have 3.

The data from operational database are raw and the data on data mart are transformed. For example, on one report my 2 order status are shown as unknown and the next report of my 3 order status are shown as 1, 7 and 9 (both 7 and 9 are transformed to unknown).

From a business point of view, the above reports are not acceptable. Any industry experiences on solving these types of problems? Why must the detailed information come from an operational database? Should all the reports source from data mart? Why and why not?

Thanks in advance

May Chen

=20