⚡ Quick Summary
Published by Google Cloud, this report benchmarks the reported business and financial impact of generative AI among global enterprises and focuses on the emerging use of AI agents. It draws on a 16-minute online survey of 3,466 upper-level, senior executive and C-suite business leaders at enterprises with more than 100 employees and more than $10M annual revenue. Google Cloud and National Research Group conducted fieldwork from 18 April to 3 June 2025. Unless otherwise stated, the report says its statistics cover respondents using gen AI in production.
The report defines AI agents as specialised LLMs with roles, context and objectives that can plan, reason and perform tasks using data, function-call APIs and, where needed, other agents. It frames agentic maturity across three levels: simple tasks, AI agent applications and multi-agent workflows. Its core comparison is between all organisations and “agentic AI early adopters”, defined as organisations allocating at least 50% of their future AI budget to agents. The report measures value through ROI, annual revenue increase and time to market, and identifies productivity, customer experience, business growth, marketing and security as the five principal areas of reported impact. It concludes with an AI agent ROI checklist centred on executive sponsorship, governed data access, human involvement, business cases and workforce skills.
🧩 What’s Covered
The report progresses from its survey design and agent-adoption picture to reported outcomes, investment patterns and implementation actions.
- Survey basis and headline findings: The opening summary reports that 52% of executives at organisations using gen AI also use AI agents in production, while 74% report ROI on at least one gen AI use case. It also states that data privacy and security is the leading consideration when selecting LLM providers.
- Methodology: The report identifies respondents by seniority, industry, enterprise size and market. It lists the survey fieldwork period and notes that results are drawn from global enterprises with at least $10M annual revenue.
- The agentic shift: The first chapter sets out the survey definition of an AI agent, a three-level maturity graphic, adoption rates by region, industry and organisation size, and cross-industry use cases. Customer service and experience leads at 49%, followed by marketing and security operations and cybersecurity at 46% each.
- Early-adopter comparison: Organisations classed as early adopters dedicate at least half of their future AI budgets to agents. The report compares their reported deployment scale, production experience, AI spending and ROI with the overall respondent group; 88% of early adopters report ROI now on at least one use case, versus 74% overall.
- Five reported value areas: The second chapter uses ROI, revenue increase and time to market to examine productivity, customer experience, business growth, marketing and security. It reports, for example, that 70% report improved productivity and 63% improved customer experience from gen AI.
- Illustrative agent scenarios: Examples describe analysing data in a Google Sheet, troubleshooting product issues, optimising stock, researching competitors and responding to a critical vulnerability. These pages also distinguish survey findings from figures attributed to IDC or to commissioned Forrester studies.
- Investment, sponsorship and challenges: The final analytical chapter covers funding models, priorities for adoption, C-suite sponsorship and barriers. It identifies systems integration and data security as major hurdles, and lists privacy and security, integration, and cost as the three leading LLM-provider considerations.
- ROI checklist: The closing checklist calls for executive champions, data governance and enterprise security, human-in-the-loop operation, secure governed access to enterprise systems, enterprise-wide guidelines, prioritised repeatable tasks, and internal AI education.
💡 Why it matters?
The report gives leaders a common set of reported outcome measures—ROI, revenue change and time to market—for connecting AI initiatives to business objectives. Its agent definition and maturity levels can help teams distinguish simple applications from tool-using and multi-agent workflows when deciding what requires operational oversight.
For governance and security teams, the practical relevance lies in the report's repeated link between scaling agents and secure access to enterprise systems, data governance, human involvement and enterprise-wide guidelines. It also records that executive sponsorship correlates with reported ROI, making governance, funding and executive alignment part of the implementation picture rather than separate concerns.
❓ What’s Missing
The report presents survey responses and comparative percentages, but does not provide respondent-level data, a causal evaluation showing that agents produced the reported outcomes, or detailed discussion of survey weighting, uncertainty or margins of error. Its global findings are aggregated across markets and the report notes that regional results are not adjusted or calibrated for cultural-bias impacts. The implementation checklist is high level: it calls for governance, security, human-in-the-loop operation and compliance, but does not specify control testing methods, approval workflows, incident-response procedures or mappings to particular laws or standards. Several financial examples are drawn from IDC research sponsored by Google Cloud or Forrester studies commissioned on behalf of Google.
👥 Best For
Enterprise leaders developing an AI investment case; transformation teams comparing agent deployment maturity; and AI governance, security and data leaders who need a concise account of the report's recommended foundations for scaling agents. It is also useful for teams selecting initial use cases in productivity, customer experience, marketing or security operations.
📄 Source Details
The ROI of AI 2025 How agents are unlocking the next wave of AI-driven business value is a 48-page English-language report branded Google Cloud. It states that the underlying survey was conducted by Google Cloud and National Research Group, with fieldwork from 18 April to 3 June 2025. No individual report authors, edition number, publication date or document URL are printed.
