Over the past few years, corporate investment in AI infrastructure has primarily focused on two paths: first, enabling large language models to access corporate knowledge bases; and second, enabling AI applications to generate more accurate responses based on corporate data. As a result, solutions such as RAG (Retrieval-Augmented Generation), vector databases, knowledge base Q&A, and enterprise Copilot have rapidly gained widespread adoption.

Many highly valuable business insights are not explicitly stated in a document or a specific field, but are hidden within the intricate network of relationships among customers, contracts, suppliers, products, work orders, code, equipment, accounts, transactions, organizations, knowledge, and processes.

Traditional search can only match keywords explicitly entered by users; conventional RAG primarily relies on semantic similarity in text; and BI dashboards are limited by predefined metrics and query logic. They are all good at answering questions where “the user already knows how to ask,” but they struggle to address the unknown—situations where “the user doesn’t yet know what to ask.”

Arango Contextus is an AI capability designed for “context discovery.” It is not a typical chatbot, nor is it merely a RAG retrieval tool; rather, it is a discovery-oriented AI that deeply integrates graph structures, graph neural networks (GNNs), and large language models (LLMs). Its goal is to push the capabilities of AI from “answering known questions” to “discovering unknown clues.”

I. Key Gaps in Enterprise AI: It’s Not a Lack of Data, but a Lack of a Relational Perspective

In a corporate environment, there is often no shortage of data.

CRM systems contain customer data and sales opportunities, ERP systems track orders and inventory, ITSM systems store service tickets and change requests, code repositories contain modules and dependencies, contract management systems store terms and performance data, and cybersecurity systems aggregate assets, vulnerabilities, alerts, and attack vectors.

The problem is that this data is scattered across different systems and comes in various formats. Even when companies feed it all into an AI system at once, they often simply slice it up, vectorize it, and then pass it to a large language model for question-and-answer tasks.

This approach can solve some knowledge retrieval problems, such as:

• "What are the setup steps for a certain product?"
• "How should a certain contract clause be interpreted?"
• "What communication records are there for a certain client recently?"

But when the issue escalates to:

• "Which customers are at a similar risk of churning?"
• "Which suppliers could trigger a domino effect in the supply chain?"
• "Which code modules share architectural similarities across different projects?"
• "Which vulnerabilities and asset combinations could create attack vectors?"

At this point, simple text similarity is no longer sufficient.

The root cause lies in the fact that the answers to these questions depend not only on the content of the text itself, but also on how entities are connected. What enterprise AI truly needs is not just a “knowledge base,” but a Context Data Layer that can understand the structural relationships within the business.

II. Contextus’s Core Value: Discovering What Users Haven’t Searched For

The logic behind traditional search is that users enter keywords, and the system returns matching results. The logic behind RAG is that users ask a question, the system retrieves semantically similar content, and then passes it to an LLM to generate an answer.

Contextus, on the other hand, is dedicated to solving a different kind of problem:When users don’t know what to search for and aren’t sure which entities might be related, can AI proactively identify relationships worth exploring?

That is precisely what makes Contextus unique.

It does not merely examine the textual descriptions of individual nodes, but rather analyzes each node’s position within the graph, its surrounding connections, indirect paths, and structural patterns. Using graph neural networks (GNNs), the system can identify entities that, although not directly connected, are structurally highly similar or have latent associations.

This capability has significant practical implications in business:

• Financial Risk Management: Looking beyond the surface data of a financial product, even if it is not directly exposed to high-risk assets, it may still be subject to indirect risk transmission through counterparties, asset portfolios, industry exposure, and related markets.
• Supply Chain Management: Two suppliers may have no direct business dealings with each other, yet they may rely on the same upstream raw materials, the same port, the same logistics channels, or production capacity in the same region.
• Network Security: Two assets may belong to different business systems but share similar vulnerabilities, permission paths, component dependencies, or attack vectors.
• Sales and Customer Management: Although these two clients operate in vastly different industries and vary significantly in size, they may exhibit similar patterns in their procurement processes, technology stacks, service issues, and decision-making chains.

These findings go far beyond what a simple search can achieve; they require the system to truly understand “relationships” and “structure.”

III. Technical Breakdown: How Contextus Works

Contextus's processing workflow can be broken down into four stages: graph data ingestion and embedding, BFS graph traversal, link prediction, and LLM explanation.

This transforms Contextus from an isolated algorithmic component into an AI-powered discovery assistant for business analysts, engineers, and decision-makers.

IV. Practical Significance for Businesses

From the customer’s perspective, companies adopt AI platforms not to chase technological trends, but to improve the efficiency of business decision-making, mitigate risks, reduce repetitive work, and uncover growth opportunities. The value of context-aware AI, such as Contextus, is primarily reflected in three areas:

First, identify hidden relationships across systems.

Enterprise systems are typically built in silos based on departments and processes—CRM handles sales, ERP handles orders, ITSM handles services, security platforms handle assets and alerts, and code repositories handle R&D. The boundaries between these systems often act as barriers to insight. Contextus unifies these entities by mapping them to a graph structure, enabling AI to perform analyses that transcend system boundaries rather than being confined to the data tables or document sets of a single system.

Second, help users discover “unasked questions.”

Many risks and opportunities do not manifest as clear-cut questions in their early stages. Customers won’t directly ask, “Which customer might churn next month?” nor will supply chain teams necessarily consider, “Which secondary supplier might impact the delivery of our core products?” If a system can only respond to questions that have already been asked, its value is significantly diminished. The significance of context-aware AI lies in proactively uncovering potential patterns, allowing business personnel to intervene early.

Third, translate technical findings into business terms.

GNNs, embeddings, and link prediction are meaningful to technical teams, but for business professionals, what really matters is “why this is worth paying attention to.” Using LLMs, Contextus translates structural similarities, connection paths, and latent relationships into natural language, significantly lowering the barrier to using graph analysis results and making it easier to integrate them into actual enterprise workflows.

V. Typical Business Scenarios

• 360-Degree View of the Customer: Contextus helps sales teams identify similar customers, potential renewal risks, and hidden cross-selling opportunities. Traditional CRMs primarily track customers, contacts, opportunity stages, and follow-up history, but a customer’s true status is often intricately linked to contracts, product usage, support tickets, past projects, technical architecture, and organizational relationships. Through graph-based analysis, AI can identify customers who “appear different on the surface but share similar behavioral patterns,” helping sales teams develop strategies in advance. For example, a customer’s current opportunity value may be modest, but their technology stack, procurement chain, service issues, and historical interaction patterns may closely resemble those of a previous high-value customer—clues that are nearly impossible to detect by looking at CRM fields alone.
• Financial Risk Management: Contextus can be used to identify indirect risk transmission paths between assets, accounts, counterparties, and market events. Many financial risks do not represent single-point exposures but rather propagate through multiple intermediary entities. Graph structures are inherently well-suited for representing such complex relationships, and GNNs can further uncover structurally similar risk patterns, aiding in risk early warning, linked-account analysis, the detection of anomalous transactions, and the identification of potential transmission pathways.
• Supply Chain Management: Companies can model suppliers, materials, factories, logistics routes, orders, regions, and risk events as graphs. Contextus helps identify vulnerable nodes in the supply chain, such as whether multiple critical products rely on the same upstream supplier, or whether a risk in a particular region affects multiple business lines. This is particularly important for the manufacturing, energy, automotive, electronics, and pharmaceutical industries, as their supply chains are typically characterized by multiple tiers, multiple regions, and strong interdependencies.
• Network Security: Assets, vulnerabilities, permissions, accounts, services, network connections, alerts, and attack paths are inherently well-suited for graph modeling. Contextus helps security teams uncover dispersed yet structurally similar attack surfaces or identify potential lateral movement paths. Compared to viewing individual alerts or vulnerabilities in isolation, this approach more closely mirrors the perspective of real attackers—who focus not on isolated weaknesses, but on the exploitation chains formed by combinations of vulnerabilities, permissions, assets, and network paths.
• Research and Development and Code Analysis: Contextus can identify structurally similar modules, component dependencies, and architectural patterns across different code repositories. The original text mentions that, on a RISC-V processor graph containing 27,583 nodes, 40,955 edges, and 24 node types, Contextus successfully identified structurally similar memory management units across different processor repositories. This indicates that it is not performing text retrieval, but rather mining underlying structural relationships across code repositories and architectures.

VI. Contextus and the Arango Context Data Platform

From a product positioning perspective, Contextus further reinforces Arango's narrative as a contextual data platform.

Arango does not simply position itself as a graph database or a vector search tool; rather, it is committed to unifying graphs, documents, vectors, search, and governance capabilities into a single platform.

This is crucial for enterprise AI architectures. When building AI applications, many companies tend to cobble together vector databases, graph databases, search engines, ETL tools, permission systems, and audit systems. While this patchwork architecture may suffice during the prototyping phase, long-term operation leads to numerous issues, including data synchronization, permission inconsistencies, query latency, system complexity, and high maintenance costs.

Arango argues that, rather than forcing multiple tools together into a single AI stack, it is better to manage graph structures, node attributes, vector embeddings, and retrieval capabilities on a unified multi-model platform, allowing context to be generated and invoked based on the same data foundation.

For enterprise customers, this is not just about simplifying the technology; it’s also about reducing the complexity of moving AI projects from proof of concept (PoC) to production. True enterprise AI applications must not only provide answers that “sound right,” but also deliver responses that are well-founded, traceable, compliant with authorization boundaries, and adaptable to changes in business relationships—this is precisely where the value of a unified contextual data layer lies.

VII. Implications for Enterprise AI Development

Contextus represents a significant shift in enterprise AI: from text-based question-and-answer systems to relationship-based discovery systems.

In the early stages, the focus of enterprise AI was on “enabling large models to understand business documents.” As applications have become more sophisticated, however, what customers truly need is “enabling large language models to understand the company’s business structure.” This means that AI must not only know what a given text says, but also identify which customer, product, contract, process, or risk event the text corresponds to—and how these elements are interconnected.

Therefore, when building AI applications, companies should not focus solely on the large models themselves or on the performance of vector retrieval; rather, they should examine whether the underlying contextual data has been properly organized.

If a company’s data remains isolated in documents, spreadsheets, and systems, AI’s capabilities are likely to remain limited to summarization and question-and-answer tasks; however, if business entities, relationships, permissions, historical records, and chains of evidence can be modeled in a unified way, AI can then advance to higher levels of analysis, reasoning, and discovery.

Conclusion

The value of Contextus lies not in replacing search, RAG, or BI, but in filling the gap that they all share: uncovering unknown clues hidden within relational structures.

• The search result said, "I know what I'm looking for."
• RAG replied, "I know what I want to ask."
• Contextus attempts to answer the question: “I haven’t yet realized what’s worth paying attention to.”

For enterprise AI, this capability has significant practical implications. This is because the most valuable insights are often not found in isolation within a single document, but are hidden within a complex network of connections among customers, contracts, products, processes, risks, code, assets, and the organization.

Through its context data platform and in collaboration with Contextus, Arango conveys a core message: the next phase of enterprise AI is not merely about feeding data into large language models, but about weaving enterprise data into a context network that is discoverable, interpretable, and governable. This is precisely the key path for enterprises to move from standard RAG to GraphRAG and ultimately toward AI Discovery.