Enterprise AI: Widening ROI Gap Between Advanced and Early-Stage Organizations

In enterprise AI adoption, a clear ROI gap is emerging between organizations pursuing advanced initiatives and those at earlier stages. This reality came to light in the survey "State of AI in the enterprise," conducted by Box, a cloud content management company, among 1,640 IT decision-makers in the United States, United Kingdom, France, and Japan.

Particularly striking in the survey is the rapid transformation in enterprise AI maturity over the past year. The percentage of organizations rating their AI adoption as "advanced" or "cutting-edge" surged dramatically from 8% to 64% in just one year. Conversely, those categorizing themselves as "early-stage" or "not started" plummeted from 53% to 9%. Additionally, 80% of surveyed organizations reported experiencing AI investment benefits in the form of at least 10% improvement, with more than half achieving measurable business results within six months of project approval.

However, the maturity gap directly translates to performance gaps. Among organizations reporting over 25% ROI from AI, cutting-edge organizations reached 50%, while early-stage companies achieved only 11%. Intermediate "advanced" organizations achieved 33%, and "developing" organizations 16%, showing the gap widens systematically with maturity level. Olivia Notteboom, COO of Box, points out that the driver of this gap is not technological capability but "rigor in integration and management." Advanced enterprises deploy agents—AI programs that perform tasks autonomously—in production environments and operate them in reusable forms, whereas early-stage companies remain at the level of individual experimentation.

The survey highlighted "content accessibility" as a critical factor driving performance differences. While 96% of respondent organizations acknowledged that AI agents need access to their proprietary content, only 36% reported successfully connecting agents to trusted content across multiple use cases. Notteboom emphasizes that the fundamental challenge now is content quality and security rather than model performance. When the quality of content accessible to agents is low, output quality diminishes; when security is weak, risks increase.

Establishing a content foundation delivers benefits beyond security. Data and information previously siloed by department become available to agents for cross-functional use, expanding organizational adoption. In other words, content layer development is not merely an information management issue but serves as the foundation for integrating AI into organizational operations.

The survey results reveal a shift in competitive dynamics for enterprise AI: the focus is moving from "which model to use" to "how to embed it in the organization." Advanced organizations are building three key elements: dedicated teams to deploy agents, a governance framework to manage operations, and a consistent content infrastructure. Only when these three elements align does high ROI materialize.

Going forward, a critical question is how the many organizations currently at the "developing" stage will close this gap. While technology itself is becoming widely accessible, establishing governance and content management systems within organizations requires time and organizational structures. Now that maturity differences have become visible as performance differences, the gap between organizations undertaking systematic improvements and those that do not is likely to widen further.