AI Data Strategy: Connection, Not Collection, Wins

AI Integration Stalls: Why Companies Struggle to Unlock Data’s Potential

The promise of artificial intelligence, particularly generative AI like ChatGPT, has ignited a fervent debate among business leaders: can we harness the power of these tools with the vast data reserves accumulated over decades? Despite the hype, successful deployments in production environments remain remarkably scarce, and demonstrating a clear return on investment (ROI) proves elusive. The issue isn’t a lack of AI capability – models are advancing at an astonishing pace – but a critical bottleneck in connecting AI to the data that fuels it.

The Data Disconnect: A Core Challenge

According to Seungjae Jeong, Head of Denodo Korea, the root cause lies in the fragmented nature of enterprise data. “For many organizations, data isn’t neatly stored in a single repository,” he explained at the Cloud & AI Summit 2026 in Seoul. “It’s scattered across data warehouses, data lakes, cloud platforms, and external APIs.” This dispersion creates a significant hurdle for AI initiatives.

<h3>Five Pillars of AI-Ready Data</h3>
<p>Jeong outlined five key requirements for effectively leveraging data with AI: a unified environment providing instant access to all data sources; real-time freshness to avoid learning delays; flexibility to adapt to rapidly evolving AI models; rich context to enable AI to understand business terminology; and, crucially, robust security and governance.</p>

<h3>Beyond Data Collection: The Rise of Data Fabric</h3>
<p>Traditional approaches to data integration often involve physically collecting data into a centralized system. However, this method mirrors past inefficiencies, incurring substantial costs and time delays. Moreover, it simply creates another point of data fragmentation. Gartner advocates for a shift in focus from data collection to data connection.</p>

<p>“True data operational efficiency is achieved when collection and connection are balanced,” Jeong stated. “A data fabric architecture, which logically integrates disparate data sources, empowers AI with the necessary integration, flexibility, and timeliness.”</p>

<h3>Implementing a Data Fabric: A Three-Phase Approach</h3>
<p>Building a data fabric unfolds in three stages. The first involves data virtualization, creating a single point of access to distributed data – a process that can be completed in as little as a week with the right tools. The second, and most resource-intensive, is building a semantic layer. This involves mapping data models to business terms, ensuring both humans and AI can understand the data’s meaning. The final stage is ontology expansion, defining relationships between data elements to enhance AI’s reasoning capabilities.</p>

<p>“A solid foundation of data virtualization is essential to mitigate the risks associated with the more complex data strategies in phases two and three,” Jeong emphasized. “Technology should handle the repetitive tasks of data connection, freeing up human expertise to focus on adding value and deriving insights.”</p>

<div class="extendedBlock-wrapper block-coreImage undefined">
  <figure class="wp-block-image size-large">
    <img loading="lazy" src="https://b2b-contenthub.com/wp-content/uploads/2026/03/DenodoKorea202603_1.png?w=1024" alt="DenodoKorea202603_1" class="wp-image-4151608" width="1024" height="597" />
  </figure>
  <p class="imageCredit">Foundry</p>
</div>

<h3>Denodo: A Single Access Point for Diverse Data</h3>
<p>Denodo’s data virtualization platform connects over 200 data sources – including data warehouses, APIs, SaaS applications, and object storage – into a unified access layer. It supports both structured and unstructured data.</p>

<p>Users can join Oracle tables with MS SQL tables and SAP BW objects as if they were within the same database. SQL-proficient engineers can query heterogeneous systems directly with a single query. Even unstructured data in SharePoint folders can be processed in bulk using natural language commands – for example, summarizing documents within 200 words. Vector databases are also seamlessly integrated, allowing queries that combine image recognition results with structured data, such as finding “homes under $600,000 with an island kitchen.”</p>

<p>Security and governance are also central to Denodo’s platform. Because all data access is channeled through a single point, consistent security policies can be applied regardless of the underlying data source. Access is role-based, with data masking automatically applied to sensitive information. A financial services client, for example, can ensure that branch-specific data is only visible to authorized personnel, and that sensitive information is filtered when accessed by public Large Language Models (LLMs).</p>

<div class="extendedBlock-wrapper block-coreImage undefined">
  <figure class="wp-block-image size-large">
    <img loading="lazy" src="https://b2b-contenthub.com/wp-content/uploads/2026/03/DenodoKorea202603_2.png?w=1024" alt="DenodoKorea202603_2" class="wp-image-4151609" width="1024" height="630" />
  </figure>
  <p class="imageCredit">Foundry</p>
</div>

<h3>Automating Semantic Understanding with AI</h3>
<p>Once a data virtualization foundation is in place, building a semantic layer becomes the next critical step. Connecting business terms to data fields is essential for AI to accurately interpret data. However, this task can be tedious for engineers.</p>

<p>Denodo addresses this bottleneck with AI. Leveraging accumulated data and metadata, LLMs automatically generate descriptions for each data model and field. Drafts are created overnight and reviewed by subject matter experts for final approval.</p>

<p>With a semantic layer in place, natural language queries become possible. Through Text-to-SQL or Model Context Protocol (MCP) integration, users can ask questions like, “Which promotions drove the most sales of which products last year?” and receive integrated results from disparate systems – cloud marketing, CRM, and sales data – instantly.</p>

<p>A Forrester study commissioned by Denodo confirmed the effectiveness of this approach. Clients reported a 65% faster data processing speed and a 67% reduction in data preparation time compared to traditional ETL methods, with a return on investment achieved in under six months.</p>

<p>“The smartest approach is to leverage solutions to handle tasks that can be automated, and focus human expertise on the semantic and ontological layers where judgment is paramount,” Jeong concluded. “It’s time to let technology handle the repetitive work of data connection, and empower people to focus on unlocking the true value of data.”</p>

<p><em><a href="https://www.gartner.com/en">Gartner</a> provides valuable insights into modern data management strategies. For further exploration of data fabric architectures, visit <a href="https://www.denodo.com/">Denodo's website</a>.</em></p>

<p>What challenges are *you* facing in integrating your data for AI initiatives? And how do you see the role of data virtualization evolving in the next few years?</p>

Frequently Asked Questions About Data Virtualization and AI