AI Investment Soars, But Most Firms Lack Operating Models to Scale

The AI Investment Paradox

Enterprises are pouring money into artificial intelligence, but few have built the operational foundations needed to capture sustained value. According to Accenture's latest Pulse of Change report, 86% of C-suite leaders plan to increase AI investment in 2026, and 78% now view it as a revenue growth driver. Yet only 32% report sustained, enterprise-wide AI impact.

The disconnect is stark: While roughly one-fifth of organizations say they are rebuilding processes for AI, fewer than one in ten are redesigning roles around it. The core problem is that most enterprises were built for software that provides information, not software that takes action.

Why it matters

This gap between investment and operational readiness threatens to turn AI spending into wasted capital. As AI moves from pilot projects to production systems capable of autonomous action, the lack of redesigned workflows, governance frameworks, and role clarity creates both business risk and missed opportunity. Organizations that solve the operating model challenge first will capture durable competitive advantage while others struggle to move beyond isolated use cases.

Four Principles for an AI Operating Model

Accenture's large-scale deployments with Microsoft, ServiceNow, and other partners, along with similar work from IBM Consulting, point to an emerging blueprint built on four key principles.

First, start with real workflows rather than abstract use cases. Organizations making progress focus on specific end-to-end processes like incident management or claims handling, documenting precisely where data sits, where decisions are made, and where handoffs fail. As Tom Bruss, managing director at Accenture, put it: "If you're just applying AI to an inefficient process, you're automating inefficiency."

Second, build an "AI spine" across data, platforms, and governance. This means creating a cloud-ready architecture with coherent data strategy and integrated workflows that support machine learning, generative AI, and agents on a common foundation. Lenovo's expanded partnership with ServiceNow exemplifies this approach, combining device telemetry with the ServiceNow AI Platform to create an integrated backbone for endpoint-driven workflows.

Third, treat governance as part of the architecture, not a separate compliance stream. ServiceNow's expanded AI Control Tower now includes a kill switch that can pause or shut down agents in real time when they operate outside their scope. This feature was developed after a real incident where an AI agent with elevated permissions deleted a production database—including backups—in nine seconds.

Fourth, keep humans in the lead and design AI-assisted work around their decisions. Accenture's internal deployment of Microsoft 365 Copilot to roughly 743,000 employees included staged rollouts, targeted coaching, and clear usage guidance by role. Internal reports indicate 97% of surveyed employees said Copilot helped them complete routine tasks up to 15 times faster, with active use rates approaching 89% in key cohorts—results that came from intentional change management, not just feature adoption.

The Role of Forward-Deployed Engineers

Deploying agents in production is fundamentally an engineering and process-design challenge. Forward-deployed engineers (FDEs) are emerging as a critical role, responsible for modernizing systems so agents can securely access data, mapping access controls, documenting processes agents can follow, and maintaining evaluation frameworks.

At ServiceNow's Knowledge 2026 conference, Accenture and ServiceNow announced a joint FDE program embedding AI-specific and industry-focused engineers inside mutual customers to build agentic workflows directly on the ServiceNow AI Platform and scale them from initial build to enterprise deployment.

Risks and Constraints

The risks are substantial. Architectural and data fragmentation remain the starting constraint for most organizations. Governance and control risks escalate quickly once agents can take production actions—a misconfigured agent can generate thousands of actions before detection.

The organizations making progress aren't avoiding these risks; they're designing with them in mind. That means narrowing scope, stabilizing critical domains first, building governed data layers, and creating control frameworks with clear scopes, central policy enforcement, and consistent logging from day one.

These details were first reported by Accenture in its Pulse of Change report and discussed at ServiceNow's Knowledge 2026 conference.