AI is no longer stuck in pilot mode — it’s delivering measurable, bottom-line impact. As organizations move beyond experimentation and into production, the conversation has shifted from “What’s possible?” to “What’s the return?”

While AI has still-unfathomed potential to amplify impact and productivity, organizations can’t realize that value until they embed AI into the ways work actually gets done. The strongest return on investment (ROI) doesn’t come from isolated pilots but from the integration of AI into core operations, powered by trusted data, strong governance and the right skills. Only this allows AI to drive sustained impact.

Our latest research underscores this idea, finding that companies investing strategically in generative and agentic AI are both accelerating at scale and obtaining a real profit.

The return on investment for generative and agentic AI is 49% — that’s $1.49 for every dollar invested — and that figure represents approximately a 20% increase over last year’s findings.

That stat comes from our new report “The ROI of Gen AI and Agents.” This global study, conducted by researchers at Omdia by Informa TechTarget, surveyed 2,050 business and IT leaders across 10 countries and reveals that enterprises are translating AI experimentation into measurable returns at scale. We’ve witnessed the shift firsthand at Snowflake, where customer discussions have steadily moved from experimentation to success in production. 

Studies in the past year have tackled the issue of generative AI and agents in different ways, sometimes focusing on failure rates of pilot projects (where some percentage of failure is required to discover success), implementation obstacles and the challenges of measuring value. This research led with two key questions: Are you seeing a return on your investment, and if you’re quantifying it, how high is the return? According to respondents, 92% are seeing a return on their AI investments.

How are these organizations achieving their success? A playbook emerges from the data in this report, and it involves a willingness to dive into this rapidly evolving technology and put it to work. This demands attention to data, which is the make-or-break factor in AI implementations.

At Snowflake, our philosophy is simple: We bring AI to the data — not the other way around. Enterprises have trusted Snowflake with their most valuable asset, and the path to ROI starts by layering AI directly on top of that governed, unified foundation. When AI runs where your data already lives — securely, with built-in role-based access controls and observability — it becomes easier to move from experimentation to production with confidence.

The evolution: From generative AI to autonomous agents

Agentic AI is moving into production faster than many expected. While generative AI creates outputs, agentic AI takes action. Though organizations may start small to prove the efficacy of their agentic solutions, agents are actively participating in today’s workflows. This shift is fundamentally changing productivity and decision-making. While many rote tasks are being accelerated by agents, humans are called upon to review, to orchestrate and to provide a level of strategic oversight that eludes the agentic solutions, at least for now.

Agentic AI marks the beginning of a true conversation with your business. For years, BI tools could tell you what happened. Agentic systems — powered by a solid data foundation — can now help explain why it happened and recommend what to do next. This shift from passive dashboards to active, intelligent decision-making is what unlocks durable enterprise value.

What’s essential for business leaders to understand is the speed with which this is happening. The shocking impact of the agentic enterprise is not the challenge of 2030. It’s a challenge being felt right now, as agentic adoption accelerates. Our research shows that 32% of early adopters already have agentic solutions in production, with another 25% planning to join them within the next year.

Crucially, these agents aren’t operating unchecked. They are being deployed as sophisticated partners to human workers, focusing on:

• Data-driven decision-making (57%)

• Improved customer experiences (54%)

• Faster innovation (51%)

Agents are also increasingly being used for software development. Nearly half (48%) of all code is now AI generated, and 82% of organizations report that agents have improved code testing and bug detection. Additionally, 80% cite gains in overall code quality. These results show how quickly agents are moving from experimentation to real, enterprise-wide impact.

Snowflake is powering this shift firsthand. With new innovations to Cortex Code, Snowflake’s AI coding agent for local development environments, developers gain secure, context-aware AI assistance directly within their preferred data engineering systems. This enables teams to work seamlessly with data wherever it lives and to build, manage and optimize production-grade workflows with greater speed and efficiency.

One of the most notable insights from the report is that executives expect roughly 41% of the agentic initiatives that they sponsor to fail over the next three years. These leaders recognize that abandoned pilots are iterations, not failures. By building this margin for error into their strategy, they are reaching production-grade solutions that drive the average reported ROI of 49% mentioned earlier.

For enterprises, this marks a turning point: The shift from generative AI to autonomous action is redefining how value is created across a business, and the leaders who operationalize it effectively will define the next phase of competitive advantage.

The data readiness gap: A reality check

Despite the optimism, a significant bottleneck remains: data silos. The survey found that:

• Only 20% of unstructured data is considered “AI ready.”

• Only 32% of structured data is prepared for AI workloads.

• 60% of organizations reported that data storage and compute costs have caused their AI projects to run over budget.

Furthermore, we are seeing the rise of “shadow AI.” Approximately 57% of respondents admit to using nonapproved AI tools. The gap is most visible in HR and sales, where many more staffers claim to use AI than IT even knows about. This underscores a desperate need for governed, enterprise-grade AI platforms that provide the tools employees crave without impacting security.

Enterprise AI cannot rely on models alone to enforce access controls or protect sensitive information. Governance must live at the data layer. When AI agents inherit existing roles and permissions automatically, organizations don’t have to reinvent security for every new AI workflow. This architectural approach prevents data leakage, reduces risk and enables responsible AI adoption at scale.

To address this issue, Snowflake unveiled Semantic View Autopilot, which automates the creation and governance of semantic views and provides AI agents with a shared understanding of business metrics to deliver consistent, trustworthy outcomes. By establishing a unified foundation, organizations can dramatically reduce hallucinations and shrink semantic model creation from days to minutes — accelerating time to value while strengthening trust.

For enterprises, solving these challenges is not simply about deploying new tools. It requires operational discipline around data readiness, cost control and governance. When these are not addressed head-on, AI initiatives will stall or sprawl. The organizations that treat data readiness as a board-level priority will be the ones that convert experimentation into durable, enterprise-wide impact.

Your strategy starts with your data

At Snowflake, we’ve always maintained that there is no AI strategy without a data strategy. The leaders reporting the highest returns are those making strategic investments in unifying their data estate. That’s because AI success is not dependent on waiting around for the next best foundational model. The models and the packaged AI solutions will be available to everyone, and they often arrive without warning. What any one enterprise can control is its own data foundation. The flashiest model won’t be much use if it isn’t running on data that is connected, governed and trusted. The model becomes commodified, while the enterprise data, unique to that organization, becomes the differentiator.

There’s also differentiation, especially in these early days, in how organizations adopt AI. The transition from generative to agentic AI represents a massive opportunity to redefine how work gets done, but models alone aren’t enough. To drive real impact, AI must be grounded in trusted, governed data and embedded into everyday workflows. 

That’s the approach behind Snowflake Intelligence and Cortex Code, which help customers put AI to work directly within their data and development environments. Snowflake Intelligence serves as the conversational front door to enterprise data — enabling business users to move beyond static dashboards and ask complex questions in natural language, grounded in governed context. Cortex Code extends that same philosophy as a Snowflake native AI coding agent that understands enterprise data and helps teams build AI-powered applications directly inside their existing environment. Together, these capabilities help organizations operationalize AI securely, quickly and at scale.

As AI becomes the operational backbone of the enterprise, durable returns will depend on more than isolated tools or model access. Organizations must unify semantic consistency, governance, cost control and agent execution on a single, enterprise-grade foundation. That’s what enables the shift from experimentation to repeatable, production-grade value with measurable profit. Snowflake’s latest innovations — from Semantic View Autopilot to Cortex Code — are designed to help organizations move from pilot to profit with confidence.

Download our full report, “The ROI of Gen AI and Agents,” to learn more about how, where and why AI is rapidly transforming the enterprise.