What Are Some Master Data
Management Architecture
Patterns?
Master Data Management will have following processes, which also demonstrate the pattern:
1.
Consolidate.
2.
Cleanse.
3.
Standardize.
4.
Govern.
5.
Share.
This is a standard cycle, which is applicable across a domain. Let me elaborate on them a bit.
Consolidate
– Data is gathered fromdifferent applications/sources which are typically named as
upstream or source systems.
Cleanse
– Once data is gathered from different applications, this will be stored at a common
place. One of the best practice or pattern one may call will be canonical data structure. One can
perform multiple operations on the collected data here:
a. Standardization.
b. Validation.
c. Enrichment.
Govern
– Once data moves into MDM application, note that above two steps are typically per-
formed outside of MDM application, data records will be matched, merged. Some new records
will be created, some will be duplicated and some may fail due to governance policies. This
needs to be reviewed by data stewards for taking further actions.
Share
– Finally, data will be shared with other applications, which will consume them.
I consider this as a standard way of Implementation. Based on this, one can create different pat-
terns of MDM architecture. I am sure, we are not discussing about the MDM application (COTS)
architecture, as this will vary depending upon the technology and vendor.
There are different styles of MDM implementation as well.
What Are Some Examples Of Bad Data (Data
Quality Issue) Commonly Seen In A Data
Management Platform?
Data Quality problems can come in the following forms:
Duplicate data
Multiple copies of the same record. MDM systems can be used to configure business rules to
detect such duplicates. Matching algorithms can be deterministic or probabilistic.
Incomplete data
Where all attributes of an entity are not available. An example would be missing zip codes in
address data
Inconsistency in data formats
An example would be a system or database where phone numbers are stored in different for-
mats like 9999999999, +1 999-999-9999, 999-999-9999, 99999 99999. Similar issues can exist
in address data wherein the addresses are not standardized (in US, this means standardized as
per USPS norms), address line 1, 2 are not filled-in properly.
In addition, there could be issues with the
accuracy
of the data itself.
What Are Some Of The Must-Have
Features In A Master Data Management
Product?
This really depends on your business need. What are the success factors for your MDM project
(not to mistake for the product(s) used in the project)? Typically, these success factors are resolved
around quite different stuff like risk mitigation, improved governance, data integration, single cus-
tomer views and what not.
Personally, I feel you should definitely judge the products by their flexibility and the way they in-
tegrate with the rest of your architecture : the reason being that MDM hubs need an awful lot of
external interaction. Second, I’d look at governance and auditing features for data, as I think track-
ing history of golden records is one of the critical things that MDM provides which a lot of other
systems don’t do, so you need MDM to do it. Third, there are more technical features like how the
deduplication or data cleansing mechanisms work. That obviously also has to match your needs,
and needs to be configurable for your various individual source systems, etc.
Page 16
Page 17
