How to Leverage the Right Data Model and Build a Robust Foundation for Enterprises

Mohit Jain • October 18, 2021
Share This Article


A good data model is like having a building plan for an architect: it provides the blueprint for any modern analytics platform or data transformation initiative. Getting the right data model in place is a key step in the database design process. A robust data model enables you to get the best performance out of your dashboards and storage, helps reduce maintenance time, and makes all the difference to your daily operations and end-user experience.

Data Modelling Approaches to Develop a Modern Analytics Platform

There are several modeling approaches, usually defined according to the standards of the organizations. To dig deeper, two of the most widely used approaches currently in the industry are dimensional modeling and data vault modeling. 

Dimensional modeling (top-down or bottom-up process) includes a set of concepts, methods, and techniques that help in designing a data warehouse. The focus is to identify, prioritize and model crucial business processes.

Case in Point: For one of the world’s largest restaurant chains, we identified a need for data transformation to better leverage at a global scale the over 50,000 restaurants in more than 150 countries and territories. We helped our client develop a common language and approach as a foundational step in obtaining value from their data assets. We enabled the consolidation of terabytes of data (approx. 500+ tables) with the integration of the source data model with a new target data model, including data dictionaries and KPI standardization which aided in building real-time Sales and Customer behavior analytics across their global markets. This, in turn, resolved organizational bottlenecks with easy access to customer data and enabled high customer engagement and retention.

Alternatively, Data Vault Modeling is an approach that is designed to support long-term storage of historical data for the long-term scenario where data comes from disparate sources. It deals with concerns regarding audit, lineage, load speed, and resilience to change. This means that every row in the data vault needs to have the record source and date attributes, which enable an auditor with data tracing.

Case in point: In Financial Services, an industry growing at a fast pace with an increase in exposure to fraud and credit risks, it is essential to foresee and minimize these risks through effective data models at scale. Hence, for a European financial services customer, we leveraged the data vault approach to build a data set that was agile enough to scale according to the requirements. It allowed for parallelization because the modeling approach had fewer points where data needed to be synchronized. This resulted in 50% faster data loading processes, a key benefit especially for the client as they were dealing with datasets in terabytes (200+ tables with delta in GBs) and handling real-time transactions or near real-time data inserts.

Case in point: For one of the largest banks in the US, we formulated a unified data platform that enables them to discover new insights and securely create business opportunities. We built an analytical layer that uses data vault modeling that resulted in a 55% reduction in data ingestion time and increase the scalability by 95% compared to their legacy system. The complexity lied in maintaining data up to date when there is frequent delta sizing in gigabytes. The new platform facilitates data consumers with autonomy to discover and share new insights at a faster rate.

The Need for Optimization in Data Modeling 

Over time, it is likely that data sources change and may result in a significant rework of the traditional models. It becomes difficult for enterprises to verify that their data is being efficiently and fully utilized to enhance business if no standards exist to check the basic accuracy, coverage, extensibility, and interpretability of data. Data-driven decisions may not bear fruit in the absence of a trusted process to maintain data quality.

Advancements in Technology That Enable Data Modeling Optimization

For businesses to stay competitive in the VUCA world, data modeling optimization will have to be conceptualized and implemented faster and more seamlessly. More and more architects and product owners will rely on emerging technologies such as graph databases to aid use case development, agile data modeling for real-time manipulation, and move towards realizing a universal data model. Some key trends that will dictate data modeling optimization include:

  • Data management on the cloud: Emerging cloud-based quantum computing promises to increase computing strength resulting in faster interconnects. The low cost and scalability will foster large applications that disseminate more data through the hardware. Development and access to data models in the cloud will become imperative.
  • Graph Database: It provides easy and quick visuals of business cases. This database type contains nodes and edges, grouping data sets, their descriptions, and relationships. They enable the evaluation of group data and a logical view of various business rules. Database administrators can scale high data values and create usable data models.
  • Agile Data Model: It provides a just-in-time data model using minimum design requirements for pertinent circumstances and deals well with a mix of data types such as relational, unstructured, dimensional, and master data. This enables business users to create their own models and eliminates the need for data engineers to provision the data, speeding up the data modeling process. Successful agile data modeling, however, requires a thorough understanding of statistics, the databases involved, the load between shared resources, use cases, the intent of data consumers, security restrictions, etc. Virtualized data platforms vis a vis adaptive data fabric enable users with more time to focus on critical business questions. 
  • Universal Data Model (UDM): It enables quicker design and deployment with improved performance and facilitates the ease of maintenance and integration at a reduced cost, in an enterprise environment. It streamlines internal communication, increases consistency of documentation, simplifies, and increases the applicability of data modeling for a multitude of use cases. 

Focus on Universal Data Model

Organizations cannot afford to rely on labor and time-intensive data models that are built from scratch. Universal data models provide data modelers with the building blocks required to develop the enterprise conceptual, logical, and physical data model with minor customizations. Each industry has its own set of business issues, with differences in personas and use cases that apply to every function. However, subject data around customers, distributors, agents, suppliers, internal organizations, and people within the organization that span across sales, marketing, customer service, purchasing, shipping, invoicing, budgeting, and human resources is identical and critical to track. 

Pre-defined templates with mainstream relationships and interactions between functions or a UDM can provide an effective solution to maintain this high-value information and ensure the data integrity issues are not ignored. A UDM provides an accurate view of unified data for business decision-making responsible for transforming data management to an enabler of a data-centric enterprise by empowering data modelers with a framework leading to high-quality designs.  

Some of the key approaches in the industry driving towards the concept of UDM are:

· Cross-Enterprise Model: Exchanging data between organizations is increasingly adopted for strategic alliances, mergers & acquisitions, etc. The Federated MDM approach is well known for the real-time enablement of these capabilities.

· Industry model: They predefine the data in terms of the business objects, business entities, business metrics, and you map the data to those business metrics. These pre-built models reduce the management effort and lead to swift implementation and ease of use.

These approaches widen the value of data modeling over singleton use cases to facilitate mainstream methods like mapping, schema, time-series analysis, and terminology standards across and between enterprises.

How Advances in Technology Enable the Universal Data Model

Entity Modeling: A basic requisite for spanning data models across the enterprise, industry-specific deployments, or between organizations is to center them on entities that are a business’ primary concern. Coupling event schema with individual entities offers advantages such as simplicity, feature generation for machine learning, and tracking customer journeys.

Federated Master Data Management: Data models expand their value via a federated approach between multiple organizations for data-supported truths by offering the exactness of exchangeable data between organizations. They enable real-time responses to changing business conditions and offer rich predictive analytics to enable collaboration for problem-solving.

Terminology Standards: Universal data models must standardize the terminology describing business concepts, especially across different data types. This will drive adoption throughout the organization.

Mapping Automation: The source mapping is separate compared to the business rules running on top. This characteristic is imperative to the reusability of common data models for the long term since the business rules don’t change with changes in the data source(s).

Time Series Analysis: UDM reduces the time spent engineering data and enables time-sensitive requests with low latency. The event-based schema exemplifies these temporal benefits with applicability throughout the enterprise. For example, events include start/stop times of client contact center interaction and sub-events such as thoughts about products, service cancellations, etc.

Case in point: We partnered with a global Bio Pharma company to help them realize their vision of accelerating digital adoption into becoming a cloud-first customer-centric organization in the industry. We built a solution capable of leveraging industrial-grade models with an analytical layer having terabytes of data augmented by an automated data-driven supply chain engine to gather insights for decision making. It resulted in 4X faster processing of clinical data leading to faster outcomes for scientists.

Sticking to the best data modeling practices is essential to prevent inefficiencies and delays. We delve into the best practices for the optimization of data models in the next section.

Best Practices for Data Model Optimization

  • Always try to go for the most granular level of data that is as per the analytics use cases for visualizations and create custom datasets.
  • Ingest only the data you need, as unnecessary chunks lead to slowed data processing. Also, using only the relevant data helps with storage efficiency.
  • Use appropriate data types and data sizes. Use numerical values whenever you can. Using text consumes more storage space and is slower when evaluated.
  • The data model needs to be accommodative of the change data capture.
  • While deciding on the model, the focus should be also on the data size, the frequency of data update, and the size of data inserts. 
  • Optimize time frames and activity days. Enable users to refresh data whenever it’s needed. This offers a better-personalized experience.

We’ve covered various approaches to data modeling, trends, key drivers, advancements in technology, and best practices for the optimization of a data model. We also discussed a few instances where we have implemented them for use cases driving successful outcomes for our customers. In a nutshell, to build a robust foundation for a business, it becomes imperative for enterprises to leverage the right data model that serves the right purpose.

Acknowledgment: This article is co-authored by Prachi Gohil and Yamini Mishra.

Let’s create something amazing together!

Contact us Next
Latest Blog
LinkedIn Instagram Facebook Twitter