Data Quality Engineering in Financial Services

Chapter 1. Thinking Like a Manufacturer

This chapter introduces concepts of efficiency, effectiveness, and Lean manufacturing along with several examples of high-quality excellence in manufacturing. Manufacturers use raw and semifinished materials to create products. Thinking like a manufacturer means thinking about data as a raw or semifinished material and thinking about data management processes much like manufacturers think about the production process in a factory. There are many similarities between manufacturing a product and managing data:

• Both use raw and semifinished materials

• Both require detailed manufacturing control specifications

• Both include quality validations and verifications

• Both use quantitative tolerances and measures to confirm conformance to a specification

Operational Efficiency

There are many definitions of operational efficiency, but I define it as the ratio between the inputs to run a business operation and the outputs gained from its production. Improving a business’s operational efficiency means the output-to-input ratio also improves. Common business inputs typically include money, intellectual property, and employees. The business outputs typically realized include products, revenue, customers, market differentiation, productivity, innovation, and so on.

Poor data quality is one of the main contributors to operational inefficiency in the financial industry. It impacts a financial firm’s ability to efficiently conduct business and it can lead to inaccurate business insights, incorrect financial analyses, and erroneous investment decisions. Further, incorrect or misaligned data negatively impacts operational efficiency whereby employees are constantly spending their time checking and rechecking data quality. Misrepresentation of financials to clients, regulators, and auditors can have severe impacts on operational effectiveness.

A firm should strive to maximize efficiency and effectiveness by ensuring only high-quality data and information are used in production data processes. Applying a manufacturing approach to data management—with precise, pre-use data quality validations—contributes to operational efficiency and effectiveness. With pre-use data quality validations, the quality of the data is confirmed, verified, and validated relative to DQS before the data is used by the consumer.

Generally, the following activities lead to a more disciplined data management operation and higher quality data used across the firm:

• Applying a manufacturing approach to data quality management

• Implementing data quality specifications (DQS)

• Implementing pre-use data validations as primary data quality controls

• Transforming reconciliations into post-use data verifications as secondary data quality controls

• Measuring the data quality of the dimensions of your data

Lessons from Lean Manufacturing

The term Lean was used to describe Toyota’s car manufacturing business during the late 1980s. A research team, headed by Jim Womack, PhD, at MIT’s International Motor Vehicle Program, began using this term in the context of manufacturing.

A Lean organization understands the essence of customer value, focuses key processes to continuously increase customer value, and creates and provides value through its processes with a zero-waste objective. A Lean organization changes its focus from optimizing separate technologies, assets, and departments to optimizing the horizontal flow of products and services through value-generating processes across technologies, assets, and departments.

A major goal of a Lean organization is to eliminate waste throughout its processes, instead of at isolated points. It endeavors to create production processes that require less capital and less human effort, and that yield fewer defects, ultimately at far less cost. The lesson from Lean manufacturing applied to data directly relates to data quality. Lean principles applied in manufacturing or an assembly line focus on eliminating waste. Similarly, thinking like a Lean manufacturer when managing data means reducing or eliminating poor-quality data that does not satisfy DQS.

Coca-Cola: Excellence in Manufacturing Quality

Coca-Cola is one of the largest multinational beverage producers in the world. The quality of the finished product is immensely important to satisfy consumers. However, in addition to consistently delivering the exact flavor of a beverage, the company must also apply stringent quality controls to all aspects of its manufacturing process. This is especially true for the raw materials such as flavoring, water, and containers since the beverage is human consumable. Coca-Cola provides a glimpse into the details of its quality assurance framework on its website, which I also summarize in this section.

Coca-Cola delivers high-quality beverage products and meets consumer expectations using precise quality measurements for ingredients, packaging, manufacturing, bottling, and distribution. All products manufactured by Coca-Cola pass through quality inspections before being released for distribution.

Quality control specialists initially check the manufacturing line before starting up the bottling process. They inspect and verify the CO₂ volumes and confirm the ratio of water-to-syrup matches the quality specifications for production. Within 30 minutes of starting the bottling process, they check the net contents to verify the proper volume is being bottled. Finally, they perform torque checks to verify the tightness of the bottle caps, and to verify that the labels on the bottles meet production guidelines.

Coca-Cola is an example of excellence in manufacturing. We generally trust in the quality and safety of its beverages because simply mixing carbon dioxide, water, and flavored syrup is not enough. Coca-Cola adheres to industry safety standards and conducts many checks and validations to ensure the quality of its finished products.

DASANI®: Purifying Water

Coca-Cola owns DASANI®, one of the world’s largest purified water producers in the world. The company uses a series of sophisticated quality controls to produce potable purified water. The purification steps performed are outlined on its website and generally include the following:

1. Volatile organic compounds and chlorine are absorbed by activated carbon filtration.

2. Minerals and additional impurities are removed by reverse osmosis.

3. Interim ultraviolet light disinfection destroys microorganisms and ensures water safety and purity.

4. Water is remineralized by the addition of small amounts of magnesium sulfate, potassium chloride, and salt to ensure consistent taste.

5. Final purification takes place using ozonation. Ozone gas, which has disinfectant properties, is pumped through the water. Because ozone, O₃, is a type of oxygen, it quickly dissipates into the same type of oxygen gas we breathe, O₂, and does not leave any residual taste in the water.

6. All steps are continually monitored and tested on a regular basis.

Refer to the DASANI® Annual Analysis Example for more detailed information on the multitude of tests used to confirm that DASANI® is in compliance with water quality standards.

These two examples illustrate the high degree of quality specifications required to produce what seem to be simple beverages.

Manufacturing Control Specifications

The production of products using labor, machines, tools, chemical and biological processing, or formulation is referred to as manufacturing. Transformation of raw materials into finished products at scale is called industrial manufacturing. The research function in manufacturing focuses on improvements and innovation in both the products and the product manufacturing processes. The engineering function focuses on product design, materials specifications, and the manufacturing transformation processes. The product manufacturing processes are controlled using quality assurance plans, control specifications, and control plans.

Water Quality Specifications

Water is used in many ways, from human consumption to the production of semiconductors. Let’s look at an example that illustrates dramatically different water quality specifications, and associated tolerances, for three different use cases: ultrapure water, potable water, and mineral water.

Semiconductor chip production requires ultrapure water. Ultrapure water, in general, only contains H₂O, a balance of hydrogen and oxygen ions, and has been purified to high levels of quality specification. Contaminants in the water used for chip manufacturing would render the production process useless.

Water for human consumption, by contrast, must meet different quality standards. Potable water is drinking water that is filtered and treated to quality specifications for human consumption. Biological pathogens, organic and inorganic matter, and chemical contaminants in the water are removed or may exist per standards at levels below maximum safe tolerances.

Potable water does not mean the water tastes good, but rather that it is safe to drink. The perception of good-tasting potable water has more to do with the dissolved minerals and human taste buds. Generally, mineral water seems to taste better to humans and often contains some combination of dissolved sodium, potassium, chloride, bicarbonate, sulfate, calcium, and magnesium. Quality in this context is defined by additional characteristics in addition to the requirements for water to be potable.

Note

The key concepts illustrated by these three water quality specifications are as follows:

• The same raw material can be used by more than one type of consumer and for more than one purpose.
• The quality specification of the material is defined by the consumer use case requirements.
• The quality specifications may differ dramatically across multiple consumer use cases.
• The quality of the material must satisfy the quality specification to be viable for use.

Quality Control and Anomaly Detection

Manufacturers use precise material and manufacturing control specifications. They apply quality tests to the raw and semifinished materials entering the manufacturing assembly line, and use sensors that measure tolerances of product components to assess their viability and to control production processes. In general, materials must successfully pass the quality gates before being moved to the next processing step.

To identify unexpected events in the manufacturing process, manufacturers use anomaly detection techniques, which employ signals generated from sensors and tolerance measurements. Quality control is a critical function that focuses on identifying anomalies and understanding their implication in the manufacturing process. This is a highly sophisticated and complex manufacturing discipline. For more on anomaly detection, consult Anomaly Detection for Monitoring by Preetam Jinka and Baron Schwartz (O’Reilly). For a detailed presentation of manufacturing-related metrics and analytical measurement models, see Smart Process Plants by Miguel J. Bagajewicz (McGraw Hill).

Summary

A manufacturing production process uses quality and control specifications to precisely define and specify the physical characteristics and quality tolerance requirements of raw and semifinished materials. The same raw materials can be used for multiple manufacturing purposes. The quality specifications of the materials are defined by the consumer use case requirements. Finally, the quality of the materials must satisfy the quality specifications for it to be viable in the production process. Manufacturing uses control specifications much like the financial industry uses DQS to engineer data quality validations.

Like product manufacturing, data manufacturing and data quality are controlled using precise DQS. Data quality is assessed before the data is provisioned to downstream processes, applications, or consumers. DQS will differ depending on the different consumer use cases. Pre-use data validations prevent data that does not satisfy DQS from polluting the downstream data ecosystem.

The next chapter introduces you to the shape of data and data dimensions. It provides the conceptual framework to understand the techniques used to measure those dimensions, relative to tolerances defined in DQS. You will see how commonly used datasets in the financial industry are data volumes, composed of panel data or cross-section time series datasets. Data volumes contain records of data that include data elements or columns. Each data element is a set of individual datum. You will be able to measure the applicable dimensions of a specific datum, and for all datum in a data volume, by applying the measurement techniques and data quality tolerances defined in DQS.