Beyond Rows and Columns: How Graph AI is Changing Business Insight by Lee O’Brien

For most of the modern business era, business data has been organised in a familiar way: rows and columns.

Spreadsheets, relational databases, dashboards, and reporting tools all share the same assumption – that the world can be described as records organised neatly into tables. A customer sits in one row, a product in another, and transactions connect them through keys and joins. This structure has powered a generation of analytics. It helps organisations track revenue, manage operations, and monitor performance.

But there is a quiet limitation hidden inside this familiar model. Businesses themselves do not behave like spreadsheets. They behave like networks.

Customers influence other customers. Suppliers rely on other suppliers. Employees collaborate across teams. Systems connect to other systems. Over time, these interactions form dense webs of relationships that shape how organisations actually function.

Traditional tables capture the pieces of information, but they rarely reveal how those pieces connect. And increasingly, those connections are where the most valuable insights live.

While most businesses were perfecting pivot tables, technology companies were quietly mapping the relationships between everything.

“While most businesses were perfecting pivot tables, technology companies were quietly mapping the relationships between everything.”

Figure 1 – From Business Data to Graph AI Insight – Modern graph-based analytics connects data from multiple business systems into a network structure. Graph Neural Networks then learn from both the attributes of entities and the relationships between them, enabling insights such as risk detection, customer behaviour analysis, and supply-chain visibility lineage for regulatory audit.

WHY BUSINESSES ARE REALLY NETWORKS

To understand the importance of graph thinking, consider how organisations operate in practice. A single invoice might connect a customer, a product, a supplier, a logistics provider, and a payment system. A purchasing decision may be influenced by industry peers, geographic clusters, or supplier relationships. Operational processes often depend on informal collaborations between teams that rarely appear in organisational charts. Each of these interactions creates links between entities. Over time, these links form networks.

A useful analogy is a city map. Imagine trying to understand a city using only a list of street names. The list tells you the streets exist, but it reveals nothing about how they connect. Without a map, traffic patterns and neighbourhood structures remain invisible.

Business data works in much the same way. Tables show individual elements of information, but they often hide the structure of relationships connecting them. Graph models make these relationships explicit. Instead of representing data purely as rows, graphs represent it as nodes (entities) connected by edges (relationships). Analysts can then explore how entities interact within a system rather than analysing each piece of data in isolation. Once those connections become visible, patterns that previously seemed random often begin to make sense.

THE INSIGHT BIG TECH DISCOVERED EARLY

Large technology companies recognised the power of connected data long before the concept entered mainstream business analytics.

Search engines offer one of the earliest examples. Ranking web pages is not just about analysing content – it also involves analysing how pages link to each other. Those links form a network, and the structure of that network helps determine which pages matter most. Social media platforms operate almost entirely as networks.

Users connect to friends, communities, and shared interests. These connections reveal powerful signals about influence and behaviour. E-commerce platforms rely on similar principles. Recommendation systems analyse networks connecting customers, products, purchases, and browsing behaviour to predict what users might want next.

Across these examples, one principle appears repeatedly:

Connections contain valuable signals.

While much of the business world focused on dashboards and reports, technology companies were building systems designed to analyse relationships. The internet itself hinted at this idea from the start. After all, it is called the web.

For years, however, this type of analysis remained largely confined to companies with massive engineering resources.

That is beginning to change.

WHERE TRADITIONAL MACHINE LEARNING FALLS SHORT

Most modern machine learning systems evolved within the same tabular framework as traditional analytics. Datasets are organised as rows, where each record represents an independent observation. Models analyse attributes associated with those records – purchase values, product categories, or customer demographics – and learn patterns from those features.

This approach works extremely well for many problems. Retailers forecast demand. Banks estimate credit risk. Manufacturers predict equipment failures. Marketing teams analyse customer segments.

But some systems behave differently. In many real-world environments, outcomes depend not only on individual attributes but also on interactions between connected entities. Supply chains operate as networks of dependencies. A disruption affecting one supplier can ripple through multiple partners before affecting production. Customer behaviour often emerges in clusters. Groups of related organisations may adopt products or services together.

in similar ways. Operational bottlenecks frequently arise from interactions between teams or systems rather than isolated failures.

These dynamics are difficult to capture when each record is treated independently. It is a bit like analysing a football team by studying each player separately without watching how they play together. The relationships matter.

WHEN MACHINE LEARNING MEETS NETWORKS

Graph neural networks (GNNs) represent one of the most important advances in machine learning, designed to capture these relationships. Traditional models analyse records independently. Graph neural networks allow models to learn from both the attributes of entities and the network structure connecting them – effectively allowing information to flow through the network during learning.

The idea is surprisingly intuitive. Each entity gathers information not only from its own attributes but also from neighbouring entities. As this information flows through the network, the model gradually builds richer representations reflecting both local characteristics and broader context.

A helpful analogy is social influence. To understand a person’s behaviour, you would examine their preferences – but also the people they interact with. Over time, individuals often adopt behaviours influenced by their networks.

Graph neural networks apply the same principle to machine learning. Instead of analysing isolated data points, models incorporate context. Graph Neural Networks allow machine learning models to learn not only from data points, but from the relationships that connect them.

WHY GRAPH THINKING MATTERS FOR SMALLER BUSINESSES

Graph analytics first gained attention within large technology firms, but the problems it addresses are not unique to global platforms. Small and medium-sized enterprises face many of the same structural challenges. The difference is scale – not structure. Understanding how customers, suppliers, employees, and systems connect can reveal insights that traditional analytics may overlook.

Case Study: Seeing Hidden Supply Chain Dependencies

A mid-sized manufacturer may believe its supply chain is relatively simple: a handful of suppliers, a logistics partner, and several distributors. In reality, supply chains rarely operate in straight lines.

When supplier relationships are mapped as a graph, a more complex picture often emerges. Multiple suppliers may rely on the same upstream manufacturer. Independent logistics providers may depend on the same transport hub. Raw materials sourced from different vendors may originate from a single producer.

These hidden overlaps create vulnerabilities. A disruption affecting one upstream node can cascade through several downstream partners, affecting production far more broadly than expected. Graph analysis helps organisations identify these dependencies before they become operational risks.

Case Study: When Customer Behaviour Spreads

Customer behaviour is often analysed individually: purchase frequency, order value, or demographics. But behaviour rarely develops in isolation. Customers influence each other through shared contexts such as industries, geographic regions, or supplier networks. A subscription software provider, for example, may notice clusters of organisations adopting a new product within the same sector.

Traditional analytics may treat these as independent decisions. Graph analysis can reveal the relationships connecting these organisations – partnerships, supplier networks, or professional communities. Once those connections become visible, the spread of adoption becomes easier to understand. Instead of targeting customers individually, businesses can identify influential clusters within their ecosystem.

Case Study: Mapping How Work Actually Happens

Inside organisations, processes are typically documented through workflow diagrams and organisational charts. In practice, work rarely follows these tidy structures. Projects often depend on informal collaboration between teams. Certain employees become knowledge hubs linking departments. Some internal systems quietly support multiple operational processes.

Mapping these interactions as a graph often reveals hidden structures. A seemingly minor system may turn out to support several critical workflows. A small group of employees may emerge as central connectors linking otherwise separate teams. For SMEs in particular – where roles often overlap – understanding these operational networks can help prevent bottlenecks as organisations grow.

THE REALITY OF SME DATA

Another challenge SMEs face is data fragmentation. Information often lives across many systems: accounting platforms, CRM tools, operational databases, spreadsheets, and third-party applications. Each system provides a partial view of the organisation. Integrating these datasets into a unified analytical model can be complex.

Traditional approaches often require heavy data engineering projects involving schema alignment and ETL pipelines. Graph models provide a more flexible alternative. Entities from different systems can be connected through relationships rather than forced into a single rigid structure. A customer in a CRM system can link to invoices in an accounting platform, which in turn connect to suppliers, transactions, or operational data. This incremental approach allows organisations to build connected views of their data without redesigning their entire infrastructure.

Graph neural networks allow machine learning models to learn not only from data points, but from the relationships that connect them.

WHY GRAPH AI IS SUDDENLY ACCESSIBLE

For many years, graph analytics was considered the domain of research labs and large technology companies. Graph processing tools were immature, machine learning methods for networks were still emerging, and the infrastructure required to analyse large graphs was expensive.

Over the past decade, however, the landscape has changed dramatically. Open-source graph libraries, modern machine learning frameworks, and scalable computing infrastructure have made graph-based analytics far more accessible. Today, organisations can experiment with graph techniques using tools integrated into mainstream data science environments. SMEs may not have the engineering budgets of global technology companies, but they increasingly have access to the same analytical capabilities.

THREE QUESTIONS LEADERS SHOULD ASK ABOUT THEIR DATA

As graph approaches become more accessible, leaders do not need to become experts in graph theory to begin exploring them. But they can start by asking better questions about their data.

Are We Analysing Data Points – or Relationships?

Most organisations analyse individual records through dashboards and reports. Yet many important signals appear in the connections between entities rather than within the records themselves. Understanding those relationships can reveal patterns that traditional analytics misses.

Where Are Our Hidden Dependencies?

Every organisation relies on networks of relationships.

Suppliers depend on other suppliers. Teams depend on shared systems. Operational processes depend on key individuals. Mapping these dependencies helps organisations identify vulnerabilities before they become operational problems.

Are We Seeing the Whole System?

Business data is often distributed across many platforms. Insights frequently emerge between those systems, not within them. Graph approaches help connect fragmented datasets into a broader picture of how an organisation operates.

FROM DATA ANALYSIS TO SYSTEM UNDERSTANDING

Ultimately, the most important shift may be conceptual.

For decades, analytics focused on datasets – collections of records describing transactions or events. But businesses operate within systems of interaction. Customers influence customers. Suppliers depend on other suppliers. Teams collaborate across departments. Systems integrate with other systems. Understanding these systems requires looking beyond individual records to the relationships connecting them.

Graph models provide tools for analysing these relationships. They do not replace traditional analytics. Tables, dashboards, and statistical models remain essential. What they add is another perspective – one designed for analysing interconnected systems. And in an increasingly connected world, that perspective is becoming more valuable.

Data tells you what happened. Networks often tell you why.