Size isn’t the problem: 3 ways to gain real insight from your data

Pundits have been saying it for years: Data is the new oil. And who can argue? Data has become an indispensable natural resource for modern enterprises, a must-have for business decision-making.

But there’s a fly in the ointment (or in this case, the oil). Organizations may be gathering data from all angles — every person, place or thing in a seemingly infinite digital trail — but to extract value, businesses must be able to answer a critical question: What is the data trying to say?

Yearning for answers, many organizations pump more and more data into storage, as if simply amassing more data into ever-growing data lakes can provide deeper insights. Yet they still end up stumped, groping in the dark for the “aha!” moments that create a greater understanding of customers, operational efficiencies and other competitive advantages.

That’s because the problem isn’t the size of data; it’s the ability to get valuable insights out of it. Business questions that help sketch out the shape of personalized product recommendations, real-time fraud detection, and medical care pathways, to name a few examples, don’t fit into the rigid way data is stored.  

Not just storing facts

Traditional systems such as data warehouses are built on relational databases (RDMBS) that are designed to store facts, not analyze data from the point of view of whom and where it came from. By nature, tables in RDBMS exist as independent files in a data lake. You may be able to find some isolated insights in that information but be blind to the insights within data that allow companies to tackle business problems with nuance.

Too often inside companies, different data points live in different organizational silos, such as sales, marketing, customer service and supply chain. That leaves a disconnected, myopic view of how an entity interacts with the business.

Even artificial intelligence (AI) and machine learning (ML) programs tend to work in silos, with each team working on a narrowly defined question. They might find answers in time, but because they’re working on separate data, they’re unlikely to discover any deeper insights (that is, patterns or similarities) that improve their model’s accuracy in answering the business questions.

Missing the meaning in data is a losing proposition at a time when organizations are under relentless pressure to gain better insights into customer behaviors, predict market changes and forecast what’s next for the business in a volatile world.

And the importance goes beyond those business uses — it also is critical for uncovering financial fraud, personalizing patient medical care, managing intricate supply chains and uncovering security risks.

Organizations have their work cut out for them in reaching an optimal state in the data journey: Uncovering the relationships within, between and among all of this information to gain meaningful insights.

How can an organization get there? Here are three key pieces of advice.

1. Eliminate silos

Many companies spend millions hiring data scientists, building new data models and exploring AI and ML approaches. The problem? These programs often work in silos across big organizations. The result? Being forced to make critical business decisions with one-dimensional data void of essential context.

Take, for example, an ecommerce company we work with that manages five individually branded retail websites. Understanding customer identities and activities across those brands is complicated, and, without a consolidated view of customer identities and activity, the company struggled to make personalized recommendations and offers.

With a new approach that traversed all the company’s customer data and synced customer identities via their mobile phone numbers, email addresses, devices, addresses, credit cards and more, the company now has a single, unified view of each buyer relationship. As a result, the company anticipates a 17.6% sales increase through its specialty retail brands.

This is a powerful example of how businesses so often gather data from disparate sources, angles and locations and store the information in silos and how that interrupts the patterns of relationships with that entity.

By merging data from different silos into one enterprise-wide dataset, companies can then analyze how a person or place or thing interacts across the business from the entity’s point of view. What is that technology? See point 2.

2. Choose the right database technology for the right workload

Relational databases, despite their name, struggle by themselves to uncover data relationships between, within and among different data elements.

Higher-level questions such as how to personalize product recommendations for customers or make supply chains more efficient require finding context, connections and relationships in data. Think about how our brains collect and store facts, data and pieces of information every second, and how the reasoning part of our brain kicks in to evaluate context and highlight relationships. 

Graph databases are a newer technology that represents an entirely different way to structure data around relationships. They act as the reasoning part of the brain for large, complex datasets for large and complex interrelated sets of data. It is within these datasets that one can see all the relationships and connections between data. LinkedIn and Meta (Facebook), for example, rely on graph databases to uncover how different users are related, helping them connect with relevant contacts and content.

By augmenting their systems with graph analytics, organizations can focus on answering relationship-based questions.

3. Unlock smarter insights at scale with machine learning on connected data

By accelerating the development of graph-enhanced machine learning, organizations can use the added insight from connected data and graph features for better predictions. Thanks to the accurate predictive power stemming from unique graph features and graph models, organizations can unlock even more potent insights and business impact.

Users can easily train graph neural networks without needing a powerful machine, thanks to built-in capabilities like distributed storage and massively parallel processing as well as graph-based partitioning to generate training/validation/test graph datasets. The result: better representations of data in terms of dealing with data type, establishing a unified data model, and having a way to represent data to get the most effective business outcomes from AI.

As these three pieces of advice show, it’s vital for organizations to adopt a modern approach to data that allows them to understand not only the individual data points but the relationships and dependencies among all data connections. To win with data, companies must be able to combine perspective, scale and speed. They also must be able to ask and answer critical, complex relationship-based
questions — and do it at the speed of business.

This is the only way today’s organizations can truly leverage data as the new oil.

Todd Blaschka is Chief Operating Officer at TigerGraph.