by Anastasia Zamyshlyaeva, Reltio
The understanding of relationships among customers, products, stores and locations is critical for decision-making in many organizations, and graph technology makes it easy to describe and visualize complex, many-to-many relationships and to model sophisticated customer objects with multi-valued, nested attributes.
Many off-the-shelf graph databases claim to do Facebook and LinkedIn-like relationship management. While suited for uncovering and handling relationships with necessary flexibility, building a Master Graph with consolidated, accurate data for Customer 360 solutions is not optimal either. Maintaining the single source of truth and provisioning it to other operational applications is challenging in pure graph databases.
Challenges for master data management
For the last few years organizations have relied on master data management tools to create a single source of truth for data entities, such as customers, products, organizations and suppliers.
Traditionally, these MDM tools have been built on top of relational databases, which come with their own challenges. Using relational tables is often too rigid and inflexible, making it difficult to support the needs of a dynamic business.
For example, doing the database modeling iteratively becomes tedious. Even small changes in a data model - such as making an organization name a multi-valued attribute - can result in costly data migration projects, or rely on temporary workarounds just to make things work, until the next change is made in the model.
Another challenge for traditional MDM solutions based on relational databases is the inability to manage the relationships between various data entities, such as people, products, organizations and places. Rigid data models require re-implementation anytime the business or data sources change. Introducing a new relationship in a relational database MDM is quite complicated and resource-intensive.
The power of the hybrid graph
Modern Data Management platforms with a polyglot approach, built on hybrid columnar-graph datastores, enable organizations to create data-driven Customer 360 applications on a reliable data foundation with a complete understanding of the relationships.
A hybrid graph-columnar storage provides an agile approach, where it is not necessary to worry whether there are one-to-many or many-to-many relationships between data entities while tackling master data reloads. This provides the flexibility to model complex relations without the burden of re-implementation when the business needs change.
The polyglot approach to customer master data also enables you to ingest omnichannel interaction and transactional data. Once you have complete and reliable customer data and an understanding of all relationships, you can use various graph analytics libraries, such as those provided by Apache Spark, to gain some interesting insights about your customers.
Here are some common use cases:
Understand customer preferences
Visualizing relationships is important, especially in retail. Consumer marketers must understand their customers very well. That includes knowing the channels they use, content they consume, their influence in social media, products they endorse and the store locations they frequent.
Consumer marketers may be hard-pressed to extract such information from traditional CRM systems. Such insights require bringing data together from all internal, external and third-party systems, including omnichannel interactions. A polyglot approach, built on columnar-graph hybrid stores is efficient at processing high volumes of transactions, good at handling queries that span the entire database and optimized to store and retrieve business entities and relationships.
Running analytics on such graph schema helps marketers understand what devices the customer uses and their channel preferences. You can learn about their product choices and send them relevant offers based on their needs.
Customer identity resolution
Determining whether a client engaging with you through different channels is the same person or not is often a challenge. The power of a graph involves creating links between incomplete data to build a complete picture of anything (for example, customers, products, places and organizations). It helps reconcile two entities describing the same person with incomplete data.
The system will compare field by field, each with slight variations and typos to find out whether the profiles are of the same individual, by matching the address, spouse’s name, children’s names and so on.
Incorporating machine learning in this process makes this even more efficient. Machine learning matching drifts away from any rule configurations with system asking questions and learning from your answers to come up with the best matching possible.
Grouping individual consumers into households
Based on an individual consumer’s social connections, locations and purchased products, organizations can group consumers into family units. In any given family, the financial buyer, the decision maker and the user may all be different individuals. The father may decide to buy books for his daughter’s birthday but may buy the books with her mother’s credit card.
Financial institutions want to understand the total worth of the household to make relevant offers rather than making decisions based on individual disconnected accounts owned by different family members. Understanding dynamics in family units using householding is important for consumer marketers.
With the help of a graph, online retailers can determine if the visitors and shoppers browsing the website are from the same household. Graph clustering algorithms can be used to infer properties or relations to link family members to one household, based on entity attributes like phone number, address, credit card information or last name.
Quick segmentation based on any attribute
Modeling data as a graph enables organizations to describe any information they have, about all domains, in a single database (customers, products, places, organizations, social media, proprietary data subscriptions) to quickly segment customers, align sales teams effectively or determine top-selling products in target segments.
Beyond graph database capabilities that just provide fast data retrieval for connected data, columnar and graph hybrid stores are optimized for large-volume analytics queries typical of data warehousing to capture relationships effectively for improved search and segmentation across any attributes.
Roll-up of dynamic hierarchical information
Understanding account hierarchies is another challenge. CRM systems cannot represent complex organization structures easily. Moreover, the customer account hierarchies available are not in the context of marketing or sales objectives. At best, you get a view of legal relationships among holding companies, business units or departments.
A graph enables you to create personalized hierarchies that can represent product penetration in a customer organization, value and credit risk assessments or revenue rollups across the hierarchy. With a personalized view of hierarchies, you can identify the business units within your customer account that are using your product, and also ones that are not. Once you identify such gaps, you will be able to target those business units to expand your coverage.
Finding key influencers
Advanced analytics running on a graph schema can help identify the most influential people in a database. By calculating scores and relationship strengths of different data elements, algorithms can be run to come up with scoring functions to find the degree of influence of a particular doctor in a particular field, for example.
A hybrid approach with a Cassandra operational data store foundation does not limit you to a single node (like some graph databases) to perform mass analytics queries across all relationships and records. It is infinitely scalable for handling master data and transactional big data volumes with ease.
Modern Data Management requires hybrid data stores designed to handle the most complex multi-domain data management challenges, and at the same time, effortlessly bring together transactional, interaction, social and machine-generated data at scale.
Obtaining a business-facing user interface to query or manage relationships, as well as maintaining data quality, requires a data store that is flexible enough to handle all disparate data types with big data scalability. With the combination of columnar and graph technologies and the power of advanced analytics, organizations can assemble a scalable data store as well as scalable analytics to deliver a new generation of data-driven applications to run a customer-centric business at big data scale.
(About the author: Anastasia Zamyshlyaeva is chief architect at Reltio, where she is responsible for the architecture of the Reltio Modern Data Management Platform as a Service (PaaS).