Three months after deployment, the enterprise AI assistant appeared to be performing well. User adoption was growing, response times were acceptable, and business stakeholders viewed the pilot as a success. The problems emerged when engineering teams began evaluating output quality at scale. Responses could not always be traced to authoritative sources, retrieval paths were inconsistent across data repositories, and observability into grounding decisions was limited. The challenge was not model performance. It was architectural. The organization lacked a reliable mechanism to ensure that every answer was consistently grounded in trusted enterprise knowledge.
For data and architecture leaders responsible for moving AI systems from experimentation to production, these numbers reveal where many deployments encounter friction. 80% of enterprise AI projects fail to deliver business value. 42% of companies scrapped most of their AI initiatives in 2025, up from 17% the year before. The single most common reason cited: their RAG could not scale past pilot. The average cost overrun for RAG projects at production scale is 380% v/s the pilot cost. And in high-stakes domains, legal AI hallucination rates still hit 69% to 88% on domain-specific questions without proper grounding. ¹
None of these outcomes is particularly surprising. They reflect the challenges organizations encounter when moving from a successful proof of concept to a production environment where retrieval quality, grounding, governance, and system observability must operate reliably at scale. Progress Software Corporation's Agentic RAG platform was built to close exactly this gap.
The RAG Cookbook, published by Progress Software Corporation, provides AI leaders, data platform teams, and enterprise architects with practical guidance for building grounded enterprise AI systems. Organizations are using it to cut hallucinations by 40%+, eliminate pilot-stage failures, and move enterprise RAG systems from PoC to production at 95% faster production readiness and 80% cost savings versus building RAG in-house. ¹
This newsletter explains why the problem is more urgent in 2026 than most enterprise technology leaders currently recognise, and what the Cookbook delivers to address it.
The Enterprise AI Grounding Gap Is a Governance and Security Problem, Not Just a Technical One
The conversation about RAG hallucinations often stays inside the AI engineering team. It should not. Many organizations initially view hallucinations as a model problem. IBM's research suggests the issue is broader than that. Once AI systems begin interacting with enterprise data, governance becomes just as important as model quality. The challenge shifts from generating answers to proving those answers can be trusted. ²
IBM's OpenRAG framework, announced at Think 2026, addresses precisely the failure mode that kills most enterprise RAG projects: fragmented enterprise knowledge. OpenRAG on watsonx. Data is an open, agentic RAG framework that connects AI to fragmented enterprise knowledge, enabling agents to search, reason, and validate, giving teams the foundation to build reliably grounded AI systems that do not confabulate answers from disconnected data sources. ³
IBM's agentic AI research further identifies the compounding complexity as systems scale: agents do not always collaborate smoothly, can compete over resources, and the more agents in a system, the more complex the collaboration becomes, with a higher chance of complications. ²
Any enterprise architect or AI platform leader who has attempted to scale agentic AI across multiple business domains recognizes this challenge. As additional data sources, retrieval pipelines, agents, and orchestration layers are introduced, complexity increases exponentially. Systems that perform well in isolated pilots often struggle when required to operate across fragmented enterprise knowledge environments. The challenge becomes less about model capability and more about ensuring reliable retrieval, consistent context construction, and governed interactions between agents, data, and enterprise workflows.
KEY FIGURES AT A GLANCE
80% of enterprise AI projects fail to deliver business value (Progress Software Corporation / The RAG Cookbook, 2026) ¹
40%+ hallucination reduction achievable with a properly grounded RAG pipeline (Progress Software Corporation / The RAG Cookbook, 2026) ¹
95% faster production readiness and 80% cost savings versus building RAG in-house (Progress Software Corporation / The RAG Cookbook, 2026) ¹
380% average cost overrun for RAG projects at production scale versus pilot cost (Progress Software Corporation / The RAG Cookbook, 2026) ¹
99% of organizations experienced at least one attack on their AI systems in the past year (Palo Alto Networks State of Cloud Security Report, December 2025) ⁴
What Progress Software Corporation's Agentic RAG Platform Is Built to Solve
One of the most common misconceptions in enterprise AI is that all RAG architectures solve the same problem. In practice, production-grade retrieval systems require far more than vector search and prompt augmentation. Enterprise teams must address data fragmentation, metadata enrichment, retrieval optimization, context assembly, evaluation pipelines, observability, and governance. The gap between a successful proof of concept and a scalable production deployment is often defined by these architectural considerations. Progress Software Corporation's Agentic RAG platform is designed to help organizations build retrieval systems that remain reliable, traceable, and measurable as complexity increases.
The RAG Cookbook gives enterprise AI leaders the practical playbook built from Progress Software Corporation's platform experience with organizations that have made that transition successfully. It covers the fundamental RAG principles that bridge retrieval and generation, the key strategies to enrich context and enhance the relevance of AI responses, the relevant metrics, including relevance and groundedness, to evaluate every stage of the RAG pipeline, advanced techniques covering multi-step queries, data augmentation, and metadata integration, and smart configuration of the RAG pipeline using Progress Software Corporation's production-tested architecture. ¹
For AI platform leaders, data teams, and enterprise architects, the challenge is rarely getting a RAG proof of concept running. The challenge is sustaining retrieval quality, governance, observability, and operational reliability as systems scale. Building production-grade RAG in-house requires engineering resources, infrastructure investment, evaluation framework development, ongoing monitoring capability, and the iterative debugging that only comes from running a system at scale against real enterprise queries. Progress Software Corporation's Agentic RAG platform delivers all of that as a governed, enterprise-ready solution, which is why organizations using it reach AI-readiness 95% faster and at 80% lower cost than teams building from scratch. ¹
Progress Software Corporation also makes the entry point practical. For organizations exploring how to operationalize RAG at scale, a two-page executive summary highlighting six key statistics shaping enterprise AI adoption is available on the same page. It provides AI leaders, data platform teams, and enterprise architects with a concise overview of the deployment, governance, and scalability challenges influencing production-ready AI systems.
IBM: RAG Is the Grounding Layer Every Enterprise AI System Needs
IBM's research on retrieval-augmented generation frames what RAG is designed to achieve and why getting it right matters so much in enterprise deployment. RAG anchors LLMs in specific knowledge backed by factual, authoritative, and current data. Compared to a generative model operating only on its training data, RAG models tend to provide more accurate answers within the context of their external data. ⁵
The qualifier at the end of that statement is the one enterprise technology leaders must carry into every RAG architecture conversation: within the context of their external data.
Retrieval systems rarely fail because they retrieve nothing. They fail because they retrieve the wrong thing, the outdated thing, or the partially relevant thing. IBM's conclusion is direct: while RAG can reduce the risk of hallucinations, it cannot make a model error-proof. ⁵
That is not an argument against RAG. It is an argument for the kind of properly configured, continuously evaluated, production-grade RAG pipeline that the platform provides as a platform. IBM's Think 2026 research confirms that by 2030, 50% of operational decision-making will be done by AI. ⁶
An AI system can only make decisions as reliably as the information supporting those decisions.
Palo Alto Networks: The Security Dimension of RAG Hallucinations
For AI platform leaders and enterprise architects, hallucinations are not simply an output-quality issue. They often expose weaknesses in retrieval governance, data controls, and the mechanisms used to ground AI systems in trusted enterprise knowledge. An enterprise RAG system that retrieves and surfaces incorrect or fabricated information at scale introduces compliance exposure, reputational liability, and, in regulated industries, direct legal risk.
Palo Alto Networks' April 2026 integration with Google Cloud addresses the governance challenge of RAG deployment directly. Prisma AIRS uses contextual grounding to prevent misleading AI outputs that contradict internal RAG data, keeping agents tied to real facts and defining safety policies that protect brand reputation and operational integrity as agentic AI systems scale. The Palo Alto Networks framing is precise: AI deployment is currently outpacing AI governance. ⁷
Palo Alto Networks' Spencer Thellmann, principal product manager for AI runtime security, frames the challenge in terms every CISO recognises: "AI is non-deterministic. You can't control what someone's going to ask your chatbot or agent, and you can't control what your chatbot or agent is going to say back to someone, which means that you have randomness on the input and the output, and that randomness means risk." ⁸
Progress Software Corporation's Agentic RAG platform is built around constraining exactly that randomness through grounding, evaluation, and production monitoring that makes AI outputs defensible.
Palo Alto Networks' State of Cloud Security Report 2025, drawing on 2,800 security leaders, found that 99% of organizations experienced at least one attack on their AI systems within the past year. ⁴
For AI and architecture leaders, an inaccurate answer is rarely an isolated output-quality problem. It can quickly create compliance concerns, erode customer trust, and undermine confidence in AI-driven workflows. In an environment where AI systems are increasingly connected to enterprise data and operational processes, maintaining grounded, trustworthy outputs becomes a foundational requirement that Progress Software Corporation's platform is designed to support.
Google Cloud: Grounding Is an Architectural Decision, Not a Configuration Setting
Google Cloud's Gemini Enterprise Agent Platform, integrated with Palo Alto Networks' Prisma AIRS at Google Cloud Next 2026, demonstrates that grounding at enterprise scale requires security and governance built into the AI architecture from the start, not applied as a post-deployment filter. ⁷
That is precisely the architectural conviction on which Progress Software Corporation's Agentic RAG platform is built.
The organizations seeing the strongest outcomes are not necessarily using the most advanced models. More often, they are the organizations that invested early in governance, retrieval quality, and evaluation discipline. They are the ones who have built the retrieval pipeline, the evaluation framework, and the governance architecture that makes LLM outputs reliably grounded in authoritative enterprise data. Google Cloud's enterprise AI deployments at The Home Depot, Walmart, and Macy's operationalise this principle at scale. ⁹
Progress Software Corporation brings the same architectural discipline to organizations that need production-grade RAG without Google-scale engineering resources.
Microsoft: The Scale of Enterprise AI Investment Demands Grounded Outputs
Microsoft's 2026 Work Trend Index, analysed by Fortune's CFO editorial team, surfaces a finding that every AI platform leader evaluating production-scale RAG deployment should internalise. Organizational factors, including culture, manager support, and performance metrics redesign, account for 67% of AI impact. ¹⁰
For AI platform leaders and enterprise architects, the question is no longer whether AI can generate a response. The more important question is whether the underlying retrieval, governance, and evaluation systems can ensure that responses remain reliable, traceable, and trustworthy when deployed in real-world environments.
Microsoft's investment in enterprise RAG architecture, including the open-sourcing of GraphRAG and its integration into Azure AI Search, reflects a consistent strategic conviction: retrieval pipelines are not enhancements to LLMs; they are the governance layer that makes LLMs safe to deploy in enterprise environments. Dynamics 365's 2026 Wave 1 release builds agentic RAG capabilities directly into supply chain, finance, and commerce workflows, acknowledging that production-grade AI requires grounded intelligence, not just generated intelligence. ¹¹
Progress Software Corporation's platform aligns directly with this architectural direction, giving enterprise teams the grounding infrastructure that Microsoft's own research identifies as the prerequisite for AI that actually delivers.
Cisco: The Infrastructure That Carries RAG at Scale Must Match Its Governance Requirements
Every RAG query, every retrieval operation, and every grounded response depends on a network and data infrastructure that can carry AI workloads reliably, securely, and without the latency that degrades both performance and user trust.
Cisco's State of AI Security 2026 report is direct about the governance gap most organizations are navigating: supply chains and enterprise workflows are growing in complexity, often without proper controls and governance, and autonomous AI agents are proliferating across critical workflows, often without accountability being ensured. ¹²
AI Security leaders tend to discover governance gaps only after something goes wrong. The challenge with AI is that those gaps can scale faster than traditional technology risks. An unmonitored RAG pipeline that surfaces ungrounded outputs in a customer-facing application, a compliance workflow, or a financial reporting function is a governance failure that no retroactive security control can undo. Cisco's AI Defense platform, expanded in February 2026, introduces AI Bill of Materials capability, providing centralised visibility and governance over every AI asset across the enterprise. ¹³
Progress Software Corporation's Agentic RAG platform is built to operate within exactly these governed enterprise architectures, with the evaluation metrics, monitoring capabilities, and audit trail that make every RAG output traceable and defensible.
What The RAG Cookbook Gives You That Your Current Approach Does Not
Most enterprise technology teams are not failing to deploy RAG because they lack ambition or budget. They are failing because they are trying to build production-grade RAG with pilot-grade architecture and are discovering the gap at production scale at a cost that averages 380% above their original estimate. ¹
The RAG Cookbook from Progress Software Corporation is the playbook that closes that gap before it opens. It is built from real deployment experience, validated against the organizations that have successfully moved from PoC to production, and structured around the five domains that determine whether an enterprise RAG project succeeds or fails: principles, context enrichment, evaluation metrics, advanced techniques, and smart pipeline configuration using Progress Software Corporation's Agentic RAG platform.
For architecture and platform teams evaluating deployment approaches, the executive summary provides a concise view of the operational, technical, and economic considerations that influence enterprise RAG adoption. Progress Software Corporation has built both the platform and the playbook because the two belong together.
The uncomfortable truth about enterprise AI is that most hallucinations are not model failures; They are information failures. Organizations already possess the knowledge needed to answer many of the questions employees, customers, and partners ask every day. The challenge is making that knowledge accessible, traceable, and trustworthy. RAG is increasingly becoming the bridge between those goals. The organizations that succeed will not necessarily have the smartest models. They will have the strongest grounding in discipline.
Download The RAG Cookbook Free Today, Presented for Progress Software Corporation.
Download Here
References
Progress Software Corporation / IntentTechInsights. The RAG Cookbook: Stop Your RAG from Hallucinating. Start Shipping Trusted AI Answers in Hours, Not Months. April 2026
IBM. What Is Agentic RAG?. November 2025
IBM. IBM Announcements at Think 2026 to Advance the Agentic Era. May 2026
Palo Alto Networks Blog. Where Cloud Security Stands Today and Where AI Breaks It. 16 December 2025
IBM. What Is RAG (Retrieval Augmented Generation)?. May 2026
IBM Think 2026. Shaping the Next Era of Agentic AI. May 2026
Palo Alto Networks Blog. Palo Alto Networks and Google Cloud Expand Strategic Collaboration to Secure AI Enterprise. 22 April 2026
BankInfoSecurity. Unpredictable by Design: The Challenges of Autonomous AI. 2 January 2026
Google Cloud Blog. Next 26: Building the Agentic Enterprise. April 2026
Fortune. What Microsoft's New Research Tells CFOs About the ROI of AI. 11 May 2026
Microsoft Dynamics 365 Blog. 2026 Release Wave 1 Plans for Microsoft Dynamics 365. 18 March 2026
Cisco Blogs. Cisco State of AI Security 2026 Report. 19 February 2026
Cisco Newsroom. Cisco Redefines Security for the Agentic Era with AI Defense Expansion and AI-Aware SASE. 10 February 2026
