April 2026 Summaries

SAP's recent API policy update has sparked significant concern among its customers and partners, particularly due to its explicit restrictions on using APIs with AI systems, including those for interaction or integration with semi-autonomous or generative AI. This policy insists that any AI-driven access to SAP data must occur through SAP-controlled pathways, raising questions about the future of data infrastructure and AI strategy for enterprises reliant on SAP. Despite reassurances from SAP's CEO about maintaining an open platform, the language of the policy remains unchanged, indicating a potential long-term impact on data integration and AI deployment. For Fivetran customers, the immediate impact is minimal as their SAP connectors remain unaffected, allowing them to continue their data operations without disruption. However, the broader implication of SAP's policy direction suggests a potential shift towards vendor-controlled data pathways, which might limit flexibility and increase costs for customers. Fivetran advocates for an Open Data Infrastructure, emphasizing the importance of maintaining control over data and AI tools, which ensures portability, governance, and freedom from vendor restrictions. As AI becomes increasingly integral to business operations, having a flexible and open data architecture will be crucial for rapid adaptation and innovation.

Open Data Infrastructure offers a transformative solution to the limitations of traditional data architectures by integrating the cost-effectiveness and scalability of data lakes with the structure and reliability of data warehouses. This approach addresses the challenges companies face with increasing data and AI demands, particularly the inefficiencies and high costs associated with older architectures that separate analytical and operational data systems. By decoupling storage from compute and utilizing open table formats like Apache Iceberg and Delta Lake, Open Data Infrastructure enhances data usability, reliability, and governance while providing flexibility in choosing compute engines based on specific needs. This interoperability reduces vendor lock-in and supports diverse analytical and operational use cases, making the infrastructure not only cost-effective but also easier to operate and adapt. As companies increasingly rely on data for AI and automation, this architecture supports scalable, trustworthy, and real-time data operations, reducing engineering burdens and enabling innovation without compromising reliability or increasing costs.

The increasing reliance on AI for decision-making has amplified the importance of data accessibility and movement, yet data silos remain a significant challenge, particularly in the SaaS industry. SaaS providers often create "walled gardens" by restricting data access and portability, which can degrade the effectiveness of analytics and AI. Although regulations like the EU Data Act grant customers rights over their data, control is often limited by platform constraints and contractual terms that favor the vendors. These restrictions are compounded by technical barriers and egress costs, as vendors control how data can be accessed and under what conditions. While customers are expected to gravitate towards more open platforms eventually, SaaS vendors currently have strong incentives to lock customers into their ecosystems. To address these challenges, customers are encouraged to choose SaaS providers carefully and consider technical solutions to ensure data accessibility, while advocating for greater transparency and pressure on vendors to change restrictive practices.

In the evolving landscape of data management, businesses once faced a dilemma between the flexibility and low cost of data lakes and the structured reliability of data warehouses, each with distinct trade-offs. Modern data lakes, however, merge the benefits of both, offering scalability, flexibility, low costs, and structured analytics without the constraints of older systems. They play a crucial role in the Open Data Infrastructure, which supports interoperability, open standards, and the integration of various data forms, making them vital for organizations increasingly relying on automation and AI. The Fivetran Managed Data Lake Service exemplifies this modern approach, facilitating efficient data ingestion and management, reducing costs, and enabling the seamless transition from traditional data warehouses to a more versatile data lake environment. This transition supports enhanced analytics, machine learning, and AI capabilities by allowing organizations to choose optimal compute engines and leverage proprietary data effectively while maintaining a robust, AI-ready foundation for future developments.

As companies increasingly integrate AI into production workflows, the challenge of data access is becoming a significant architectural constraint, shifting from periodic human-driven analysis to continuous machine-driven interaction. Traditional data architectures, designed for batch processing and analytics, are inadequate for the demands of agentic AI, which requires consistent, reliable, and interoperable data access across various systems. This has led to the need for Open Data Infrastructure, an approach that emphasizes open standards, interoperable storage layers, and decoupled system design to maintain data portability and governance without being restricted by platform-specific controls. Open Data Infrastructure allows data to be stored once and accessed across multiple environments, enabling organizations to adopt new tools and frameworks without being confined to a single vendor's ecosystem. This flexibility is crucial as AI systems increasingly require direct interaction with trusted data, necessitating a robust, adaptable architecture capable of supporting evolving workloads while ensuring data consistency and governance.

Enterprise AI initiatives often fail to deliver measurable financial impact due to underlying data issues rather than deficiencies in AI models themselves. A study by MIT highlights that 95% of generative AI pilots did not affect profit and loss statements, pointing to a "learning gap" where AI systems struggle with brittle workflows and lack contextual understanding of data. The core problem lies in poor data integration and quality, with AI agents requiring reliable, current, and consolidated data access, unlike traditional dashboards that tolerate data gaps and inconsistencies. Effective AI deployment demands robust data infrastructure, including semantic layers to provide context, and governance to ensure compliance and performance. Data quality is paramount to prevent AI systems from making confident yet incorrect decisions, and freshness of data is critical for real-time applications. Organizations that prioritize data infrastructure as a foundational aspect of their AI strategy, rather than an afterthought, are more likely to succeed in scaling AI initiatives beyond pilot phases, as the potential for AI is only as strong as the data it relies upon.

The interaction with data is rapidly evolving as AI assistants like Claude can now perform tasks such as writing SQL, interpreting results, and providing insights in natural language, traditionally requiring a full cloud warehouse. This post explores how to leverage Open Data Infrastructure (ODI) to access AI-powered data experiences directly from a laptop by connecting Claude Desktop to a Fivetran-powered data lake using DuckDB as the compute engine. ODI emphasizes storing data in open formats and decoupling compute from storage, allowing flexibility and eliminating vendor lock-in. DuckDB, an in-process SQL engine, facilitates local querying of data stored in Iceberg formats, while Claude, an AI assistant from Anthropic, enhances data interaction by interpreting query results and answering analytical questions in plain English. The Model Context Protocol (MCP) links Claude to live data without the need for data export or middleware, transforming data querying into a conversational experience. This setup empowers data teams by reducing the need for extensive SQL writing, allowing them to focus more on data interpretation and action, with data remaining in an open, shared format accessible through various platforms without duplication or migration.

Insurers transitioning from on-premise Guidewire InsuranceSuite to Guidewire Cloud aim for enhanced agility and faster upgrades, with Guidewire Cloud Data Access (CDA) playing a crucial role. CDA facilitates near real-time streaming of incremental change data into Amazon S3 buckets as Parquet files, offering a more advanced foundation than traditional batch extracts. However, the challenge lies in making this data usable for business teams, as it requires continuous ingestion, mapping, and adaptation to evolving schemas. Fivetran's Amazon S3 connector, featuring Dynamic Table Mapping, addresses these issues by automating the mapping of CDA files to destination tables, eliminating the need for manual ingestion logic and reducing engineering workload. This solution enables business teams to access timely, analytics-ready data while minimizing pipeline failures and maintenance tasks. As insurers move to the cloud, Fivetran's approach allows for a standardized, automated data ingestion process that enhances efficiency and supports high-value use cases across various departments such as claims, underwriting, and finance.

Enterprises are increasingly facing challenges in harnessing AI due to systemic issues with data integration, rather than limitations in AI models themselves. Despite significant investments in modern data warehouses and business intelligence tools, many organizations struggle with maintaining reliable data pipelines, leading to delays in AI and analytics initiatives. These delays, often measured in weeks, incur substantial costs due to downtime and lost revenue. The crux of the problem lies in outdated, DIY, or legacy data integration approaches that cannot handle the complexity and scale required by AI workloads. Successful enterprises differentiate themselves by treating data integration as scalable infrastructure, which reduces failures, speeds up recovery, and minimizes manual maintenance. This shift allows them to transition from AI experimentation to production effectively, ultimately enhancing their operational confidence and maximizing revenue potential.