Agentic AI at Scale: From Hype to ROI

Google Cloud’s 2025 ROI of AI study signals a step-change: AI agents are now in production at scale and delivering measurable business outcomes.

Adoption Shift: Pilots to Production at Scale

The study, fielded with National Research Group across 24 countries, finds 52% of executives report their organizations already use AI agents—specialized models that can plan, reason, and take actions. Momentum is material: 39% say their company has launched more than ten agents. Executives also report faster delivery cycles, with over half moving use cases from idea to production within three to six months, up from last year. Generative AI investment continues to climb as technology costs fall, and nearly half of organizations are reallocating budgets from non‑AI programs to fund AI initiatives.

Early Adopters Deliver Stronger ROI

A distinct cohort of “agentic AI early adopters” has emerged, representing roughly 13% of leaders surveyed. These organizations are embedding agents deep into operations and earmarking at least half of future AI budgets for agentic systems. Eighty‑eight percent of this group reports ROI on at least one generative AI use case, versus a 74% average across the total sample. They also see higher ROI incidence in customer service and experience, marketing effectiveness, security operations, and software development. Importantly, business growth is holding steady: among leaders who report revenue gains from generative AI, most estimate increases in the 6–10% range, consistent year over year.

AI Agents in Telecom: Business Impact

Telecommunications leaders report broad, multi-domain deployment of agentic AI to improve customer experience, harden security, and accelerate network automation.

Top Telecom Use Cases for AI Agents

Among telecom executives surveyed, 56% say their organizations use agentic AI today. Adoption spans the stack: 47% use AI for security operations and cybersecurity; 46% for tech support; 45% for customer service and experience; 45% for product innovation and design; 43% for marketing; 43% for productivity and research; 41% for software development; and 39% for network or equipment configuration and automation. Nineteen telecommunications use cases have already delivered positive ROI, with the top tier including customer service and experience, security operations and hypersecurity, tech support, marketing, productivity and research, product innovation and design, network configuration and automation, finance and accounting, software development, and network remediation.

Measured Gains in CX, Security, and Operations

Telecom respondents point to measurable gains: improved productivity (reported by 72%), faster time to insight (58%), better accuracy (55%), non‑IT process efficiency improvements (55%), and faster time to market (44%). Security outcomes stand out as well. A slight majority say generative AI has materially strengthened their security posture, with most citing higher threat identification rates, better integration of intelligence and response workflows, lower time to resolution, and fewer security tickets. At the macro level, operators rank the top five enterprise impacts as productivity, customer experience, business growth, security, and marketing.

Enterprise AI Playbook: Assistants to Multi-Agent Systems

As AI budgets scale, the decision calculus is shifting from model features to enterprise-grade privacy, security, and integration.

Data Privacy and Security as Top Criteria

When selecting large language model providers, more leaders now prioritize data privacy and security as a top-three criterion, ahead of esoteric features. That aligns with rising deployment of agents capable of taking actions across sensitive workflows. The takeaway: enterprises need modern data architecture, clear governance, and policy-driven controls for safe agent execution. This includes data minimization, role-based access, auditability, lineage tracking, and structured red-teaming for prompts, tools, and actions.

Integration and Time-to-Value Drive ROI

Scaling agents requires connecting models to enterprise systems, telemetry, and tooling. Organizations reporting the strongest ROI pair AI services with robust integration into CRM, ticketing, observability, cybersecurity tooling, and developer platforms. Shorter time-to-production cycles—now three to six months for the majority—are enabled by reusable components: orchestration frameworks, retrieval-augmented generation, tool and API catalogs, and evaluation harnesses. In telecom, alignment with TM Forum’s Open Digital Architecture, intent-based network management, and closed-loop automation helps translate agent outputs into operational actions.

Design for Multi-Agent Orchestration, Not One-Off Bots

The pattern shifting from chat assistants to systems of cooperating agents is especially compelling for telco operations. Examples include a customer-care agent grounded in policy and knowledge bases, a diagnostics agent interacting with OSS/BSS and inventory, and a remediation agent proposing changes for human approval or triggering safe automation. Coordinated evaluation, safety constraints, and human-in-the-loop checkpoints are essential, particularly where agents can affect network state or customer data.

12-Month AI Agents Action Plan for Telcos

Leaders can capitalize on the early-adopter advantage by focusing on ROI-rich use cases, hardening foundations, and industrializing delivery.

Prioritize High-ROI Telecom AI Use Cases

Start where value and feasibility intersect: contact center deflection and next-best-action in care; security operations triage and threat hunting with tool use; field service and tech support copilots; network configuration, assurance, and remediation in low-risk domains; and marketing content and offer optimization with governance. Set business KPIs upfront—AHT, FCR, MTTR, NPS, threat detection rates, churn, and opex—and run controlled A/B or champion–challenger tests.

Embed Guardrails and Governance in the AI Platform

Standardize on an AI platform that offers strong data isolation, enterprise identity, policy enforcement, and comprehensive logging. Implement prompt, tool, and action allowlists; structured approvals; red-team routines; and continuous evaluations. Embed security-by-design: PII controls, secrets hygiene, content safety, and incident playbooks. For telco operations, define clear blast radius limits and rollback strategies before enabling any write or execute permissions.

Industrialize Delivery and Track AI Economics

Create reusable components for retrieval, tool connectors, observability, and test harnesses to reduce time-to-value. Adopt cost and performance management practices for AI (token budgeting, caching, model routing, and autoscaling) to sustain ROI as usage grows. Integrate agents with developer workflows to boost software delivery and with AIOps to accelerate incident response. Finally, plan for vendor diversity across model providers while meeting enterprise requirements on privacy, security, and integration—Google Cloud’s findings underscore that these are now the deciding factors for durable returns.