Graph databases team up with BI: It’s all about relationships
Traditionally, business intelligence (BI) solutions have been built on relational databases and employ very well-established technology. They are used for reporting and for predictive analytics, identifying patterns of behavior that indicate fraud, for example. Graph databases, conversely, show connections and relationships among entities, and are used in such applications as LinkedIn and Facebook. In the past few years, methods of integrating graph database technology with BI solutions have emerged in order to leverage their combined strengths.
Although relational databases have been around for about 50 years and are the most popular type of database, complex queries are difficult and time-consuming, and the databases can also miss out on some types of critical information. “In some cases, the presence of a relationship is more critical than a numerical value,” said Khalid Morris, senior consultant in the enterprise performance group at Deloitte. “Ideally, transactional information such as purchasing behavior should be linked with other activities such as travel history. This type of analysis is much easier to do in a graph database and examine behavior more closely.” (For more on identifying relationships, see the section titled Associative engines make connections.)
Adaptive supply chain management
Jaguar Land Rover (JLR) is a U.K.-based manufacturer of high-end consumer vehicles. The company employs more than 40,000 people worldwide and has global sales of about £20 billion. Its supply chain is large and complex, relying on a base of 30,000 parts; about 4,500 are used in each vehicle. Numerous systems are required to support planning and distribution, including ERP systems, parts data, and simulation applications that allow prospective customers to configure the vehicle they want.
The supply chain can be disrupted in unanticipated ways, including natural disasters, geo-political events, and materials availability. Similarly, demand is affected by many factors, particularly economic conditions. COVID-19 has had a significant impact on both the supply chain and on consumers’ purchase plans. Responding to these rapid changes is difficult, since the impact of a change in one system affects multiple other systems in a domino effect. Adding to these challenges is the fact that manufacturers must make commitments for supplies years in advance and, if they do not meet their target orders, they are penalized.
In order to respond more quickly to unexpected events, JLR decided to incorporate a graph database into its business analytics system. Graph databases store data about the relationships among entities, providing insight into the interdependencies among them. The approach provides a method for adapting to changing supply and demand at a speed that would be impossible to achieve with a relational database. The company selected TigerGraph DB on the Google Cloud Platform (GCP) as its analytics engine.
JLR uses Google BigQuery for its data warehouses, which draw upon 12 separate data sources (the equivalent of 23 relational database tables), and uses Tableau as the interface for queries, reporting, and visualization. TigerGraph imports data from the data warehouse, then analyzes the data and delivers the output back to Tableau via Google BigQuery. Complex queries can be generated and answered more quickly because the entities are pre-connected; they do not need to be recomputed during the analysis.
The relationships are established in a schema managed by the data science team at JLR. Although the upfront investment to set up the mapping is significant, once it is done, performance in response to queries is much faster than it would be with a query made to a relational database. “This system reduced query time from weeks to less than an hour,” commented Guarav Deshpande, vice president of marketing at TigerGraph. The reason the queries took so long previously is that in order to plan the full supply chain, which require matching up the demand (orders) with supply (parts), all the queries need to complete, which is very time-consuming.
Deshpande cited the case of the Range Rover Evoque, a new compact SUV from JLR. “Demand for this popular model picked up significantly in the spring,” he noted, “which caused the company to reassess its forecasting.” TigerGraph analytics allowed the company to conduct the complex queries it needed in order to understand the implications of the change in demand. The variables included the order commitment, the availability of parts, and the possibility of diverting parts earmarked for other models but usable in the Evoque. “TigerGraph can incorporate factors such as whether a shipping border is closed, or a plant is working at only partial capacity,” he added.