Data Quality indexes in practice
In one of the previous posts, I wrote about Quality indexes and why organizations can benefit from implementing them. Today I would like to take a deeper dive into this topic with a more practical guide.
The main advantage of structuring a high-level quality index is the transparency of dataset quality and at the same time ability to drill down to lower levels for finding the root of the problem.
Let's get right into this and see how one can setup Data Quality KPIs:
1. To a start get stakeholders' input and try to understand what data are they using and how. This is key to the process of setting meaningful measures and creating compliance rules on most critical data issues:
Define Attributes that are relevant for business ( street, postal Codes, VAT number)
Perform data profiling and discover which of them are having critical, issues
Create data compliance rules for a system: Customers need to have a street, postal Codes, VAT number
2. List out key systems, objects, and processes using and creating data. For example in the CV domain we can check:
CRM system (like SalesForce) :
- reference tables with country codes or postal codes
- Master Data objects - Customers and vendors
- Transactional data - Opportunities and possibly quotations
Master Data systems acting as source of truth
- Reference data
- Master Data
ERP`s system used for orders and transactions
- reference tables with country codes or postal codes
- Master Data objects - Customers and vendors
- Transactional data - Sales and Purchase Orders
You can have indexes for each Application or system and have Application Manager accountability in place.
3. Now you can Design an index for the dataset or system by combining your inputs and assigning priorities. You can measure the quality of Customers data in ERP as:
Percentage of customers fully compliant. You are defining a critical attributes list that needs to be on tiptoe. You don't care if one or 5 of the attributes are ok you need to have all or no go this record. Then you measure index as a percentage of customers records that are in line with this setup
A weighted score of compliance for each attribute:
Define a list of attributes you are measuring: Street, Postal Code, Country.
Calculate Quality for each attribute
To get QI to assign weight to each attribute and calculate the weighted average (score for street x weight A + score for postal codes x weight B + score for VAT x weight C)
Hybrid - You combine the first two approaches by saying that the wrong VAT is a hard stop and the record score is 0 and you threat it as not usable. you calculate a score for each record and then the average is of records is QI for the data source
4. Then You can define Quality metrics on 3 objects level:
Reference data quality - if the reference table you are using when assigning attributes is up to date - for example, if SAP list containing Incoterms values or country iso codes according to standard.
Master Data quality - so data of Customer/Vendor record attributes shall be aligned with reference table or validation rules
Transactional Data Master Data usage - if transactions are valid:
transactions in local ERP's have global customer identifiers, not just local ones.
Transactions are having attributes defined in the Reference Data library
As a rule of thumb, you should start from the bottom level. Once reference tables are cleaned and Master Data objects can be created only with dropdowns instead of free text you can move to Master Data objects and once you finish with those you can take care of transactional data cleansing. The main focus should be on improving the process of data maintenance and creation instead of making the whole project a cleansing exercise
5. Quality indexes can be categorized as is Master Data in different levels depending on maturity and people accountability. It's all depends on data governance and ownership setup.
Starting from the Domain level we have areas of:
Business Partner
Products Material
HR
Finance
Location
You can have a data Governance manager role for each domain and make them accountable for the domain quality index.
Additional Steps:
6. Once you have your Compliance KPIs in place you can move forward with improvements and incorporate additional activities :
Deduplication:
Maintaining redundant data can be costly. This is are when consistency rules can help. A trustworthy and clean Master Data system can be key to doing this with ease. Once you know your data is clean of duplicates and you are forcing consistency rule you can easily see local records that need your attention as they will either miss global identifier or you'll see multiple of them mapped to one.
Consistency Rules and Data syndication:
This is a big milestone in measuring data quality, once you establish one of your systems as a source of truth. Just use mapping of the slave system to master and compare if attributes match each simple as that. It should apply only to relevant attributes as in some cases you'll still need local names or addresses. Besides finding inconsistency in Business partner data you'll also be able to discover records that are mapped to the wrong masters.
This can be very helpful in a complicated IS landscape but it can be also overprocessing trap and you should focus only on real needs from the business.
To better understand this, you should try to set up validation rules on newly created objects and see if there is an improvement in processes not only data itself.
In the end is all about ROI in your data, with a robust data creation process, you can follow right for a first-time approach which leads to lower data maintenance costs and no need for future cleansing activities.
