© 2019 SourceMedia. All rights reserved.

Data Quality in 5 Steps

As the new customer master data governance leader, I had my work cut out for me. My company, like many, was composed of separate entities whose order entry systems, processes and definitions were created independently or, over time, became unique. It was my job to make sense of this. I focused on imagining what our data could do once it was complete and trusted. It was all about enabling growth.

This story begins when the CEO attempted to obtain better terms and conditions from a difficult customer. Making what he thought would be a reasonable and simple request, he asked his leadership team for the total dollar value of the last quarterly sales to this customer. The answer would take weeks to deliver.

Customer data is a complex, moving target. My colleagues cautioned me that my job could not be done. Each business had its own way of doing things and a strong desire to stay that way. Multiple versions of SAP both between businesses and within businesses further complicated by old legacy systems and a complex distribution of customer creation authority across the global landscape seemed impossible to tame. Little did they know

Lean Principle 1: Specify Value in the Eyes of the Customer

Using a Six Sigma DMAIC (define, measure, analyze, improve, control) approach and a focus on Six Sigma Lean tools, the plan began to take shape. As DMAIC specifies, before you do anything, define the problem. Begin with a voice of the customer. Understand what the customer needs. Remember that it's not about good data, but what the business can do with trusted data from across the corporation. It is then possible to create the plan and act upon it. This will help to define your problem statement, the burning platform you can use to convince key stakeholder of the problem's urgency. 

At my company, the targeted audience for the voice of the customer was sales and marketing. Interviewing this group revealed benefits to be gained from a structure that transcended business unit boundaries. Knowledge of customer interactions would enable new opportunity identification, improve win rates from leveraging the relationship and improve customer retention. Sales and marketing could prepare for customer visits with the latest information, avoiding surprises at a customer site. Business leaders could identify the top customers at a glance by business, by region, by standard industry code and at an enterprise level. This would also help with mergers and acquisitions to reveal if a potential acquisition might be a competitor to an internal business. Reports on activity and dollars could be pulled from across the enterprise in one simple transaction. Discovery by chance could be changed to research by design. All this depended on trusted data. 

Lean Principle 2: Identify All the Steps Along the Process Chain 

I needed to develop a process to deliver continued improvement. The customer data began to come into our central repository, which is named Customer Data on Demand or CD2. Dun & Bradstreet assisted in data cleansing and enrichment with family alignment and SICs. D&B matches each customer record to a DUNS number from their database and assigns a confidence code. A tolerance level was set at a confidence code of eight out of 10. Lower confidence codes were put into the low confidence bucket, and zero confidence records were categorized as a nonmatch. Low confidence and nonmatched records needed to be sent back to the business group to determine if a DUNS number match could be found with improved data. Three important metrics were identified: 

1. Quality, determined by match rate.2. Completeness, determined by population of core fields.3. On-time-to-request (OTTR), measured by file receipt into the central repository by due date. Next, I needed to establish scope and criteria. After recruiting team members from across the enterprise, we had our first meeting in Arizona in the dead of winter. In addition, numerous entities were represented in the Phoenix area, facilitating their participation. We spent three days developing our charter and scope and defining our mission, scope, objectives and associated measures. Our immediate focus would be as follows: 

1. Define core data and the associated rule. 2. Find a tool to assist in data cleansing for those groups involved in legacy loading into SAP. 3. Improve our match rate from 64 percent to 85 percent to support our value proposition. Trusted data was key to our value proposition, which was stated as data integrity, customer alignment and increased productivity through common process, as well as reliability and validity of reporting and analytics. We defined requirements for describing trusted data: 1. Accessible: Data is independent of the originating source.2. Available: Data is there when needed and of a defined and trusted quality.3. Consistent: Definitions have been agreed to by all. 4. Auditable: The origin is traceable.5. Secure: Data is a valued corporate resource. Data ownership was a big issue. The naysayers continued to cry that our conglomerate culture prevented a coordinated effort to drive data quality. I realized that the main risk was human and identified the "what's in it for me" (WIIFM) syndrome. I offered the businesses end-to-end process control over their data. They had the purchase orders and knew the customer, sales and customer service teams, enabling them to control data quality and the subsequent match to the DUNS number. I started with a business that I believed would support the effort for a quick hit and a positive example, and I worked toward areas where I expected increasing resistance. A key enabler was my ability to provide the project management support, tools and training to enable the effort. 

I analyzed the data and presented each team with a Pareto view of transaction dollar values for their unmatched data. Within each feeder system, the subgroups were the top 100 low confidence and top 100 nonmatch records by transaction dollar value. I also checked the remaining customers for outliers to ensure that a major customer record was not missed due to the low dollar value. The vast majority of the time, top customers were also the top 100.It was essential to label needed items in order to maintain transaction traceability. The file code from the original sending system and the source system customer number were tied to the record, providing an audit trail. The DUNS number would be the system-independent identifier in addition to providing the family linkage. Completed customers would be stored within families. 

To create a place for everything (standard work), I designed an online tool. The low confidence records included the source system identifier, customer number and the customer data, including the dollar value provided and the D&B suggestion for name, address, telephone number and global ultimate (family alignment). The data steward could verify the match, mark it correct or, if not, input the true match. The nonmatch record data could be verified as correct or corrected and re-sent with corrections to D&B for another attempt at matching with better data. 

Lean Principle 3: Make the Process Flow 

At this point, it was necessary to engage the business. Data stewards were the key to improving our data. Fortunately, there are already people within every organization who do this task, just not in a coordinated effort. Members of our newly formed governance council provided leads and contacts covering multiple areas, including customer service, credit services, sales and SAP order management. In one business unit, each business line had its own system. Each of these required a different steward, and each was divided into regions, further segmenting the teams. Although some teams were immediately engaged, some took up to one and a half years to be fully involved. My hardest sell was the Asia-Pacific team. Their initial response was, "This is a corporate problem. It doesn't affect us." I reached out to their VP of financial integration, and after demonstrating the dollar impact of the unresolved records, I inquired about their scheduled SAP implementation over the next two years. Their management understood the urgency. Again, offering support was key. I volunteered to mentor a Green Belt team. The project had two parts: correct and align the low confidence and nonmatch records from the last 24 months, and create one standard process for new customer setup, including English equivalency. Finding the WIIFM is not always quick, and execution is not always easy. The team included members in Korea, Australia, China and India as well as myself at headquarters in New Jersey. We met at 6 a.m. EST, and when daylight savings time had begun in Australia but not yet in the U.S., we had several weeks of 5 a.m. meetings.

Lean Principle 4: Make Only What is Pulled by the Customer

My department purchased several licenses for the D&B online tool, Hoovers, which provides a subset of the D&B database for company research. This tool, along with their own internal customer knowledge and Web searches, empowered the data stewards to bring our data up to the defined standard at a very low cost. The data was always in a consistent format, my team and I created and conducted training for the other teams, and the corrections were mechanized through the online tool. 

The next step was to document and obtain agreements. Our data dictionary program included those customer master data elements required at a corporate level. The data governance team defined our core fields per SAP definitions choosing the Legal Name, Doing Business as Name, Street Address, Postal Code, P.O. Box and associated Postal Code (if different), City, Region as State or Province in the U.S. and Canada only and Country. In addition, the DUNS number was selected as the system-independent identifier to be loaded into the SAP FI Credit master in the Credit Information Number (KRAUS field in KNKK table-Credit Master). All of this was done using a standard Six Sigma DMAIC approach clearly defining our problem statement, quantifying risk factors and requirements to drive consensus. 

Lean Principle 5: Strive for Perfection by Continually Removing Waste

As a measure for predictability, our metric for success was the enterprise yearly revenue and percent alignment with the D&B DUNS number. Targeted initially to improve from 64 percent to 84 percent, I was able to obtain 90 percent through strong business partnering and driving ownership to the point of entry. The lesson was to always go to the source. Data quality is not owned by IT but by the data consumer in the business. Weekly meetings were required to keep all the teams on track. As many as 60 people worked the data across the span of our businesses. All of these resources were already managing the data. No new recourses were required, just a coordinated effort. 

A philosophy of getting it right the first time drives cost containment and points to the need for mechanized standards. Rework wastes the cost of employee time and the lost opportunity to work on other projects that add to growth. Using a data quality tool, data can be correct at the beginning. We chose Informatica Data Quality, which has functionality to partner with D&B to cleanse and enrich with the DUNS number as the data is created. There will always be a cleansing effort due to acquisitions, legacy system issues, etc. But after the baseline is created, tools can keep this at bay. Using a data quality tool to load data from a legacy system into the new SAP instance resulted in a 99 percent productivity improvement, allowing the team to cleanse and fix in four days what would have taken four months. Data completeness cannot be obtained without eliminating scrap and ensuring agreements for data delivery. Initially, many files were months late coming into our central repository. This was unacceptable. The technical team told me that there was nothing they could do about it because of our culture. 

Six Sigma begins with understanding. Brainstorming sessions with business system owners enabled on-time file delivery. Knowing that what is measured is managed, I developed a scorecard for scrap (unusable records) to measure our progress. We graduated from receiving records with no name, no country, no dollars, etc. to seeing positive results. One group improved their data quality more than 80 percent by enabling a single edit routine.

We began with a 64 percent revenue alignment to DUNS number and ended with a 90 percent alignment. We now had control over our data. A true cultural shift had taken place. We had clear data ownership at the point of creation. Teams understood the power of the DUNS number, our metric for quality, which enabled family alignment, revenue recovery and credit risk management, always essential, but especially in this volatile time. We engaged visual controls, measuring completeness by identifying scrap and OTTR with scorecarding posted on the site. Information sharing provided through a centralized forum for customer data management in the centralized repository and best practice sharing through the governance council trumped the old way of process failure and decision-making by conjecture.

For reprint and licensing requests for this article, click here.