Enterprise AI is delivering business insights but not the cost savings most companies expected

Enterprises are deploying AI at a faster clip than ever, but the returns are not landing where most business cases said they would. An SAP survey, covered by CIO Dive, found that AI is helping organizations surface business insights and improve customer interactions, yet cost savings and time efficiency, the two metrics most procurement and IT leaders used to justify initial investments, remain elusive for many.

That disconnect matters to any team now weighing a broader rollout of platforms from Microsoft, Salesforce, or Google. All three vendors have expanded their AI-native toolsets considerably, and competitive pressure to adopt is real. But the operational case for scaling needs to rest on what AI is actually delivering, not what vendors projected at launch.

Three platforms, three different bets

Microsoft Copilot, embedded directly into Microsoft 365, has drawn the most enterprise traction of the three, according to reporting from TechRadar AI. Its value proposition centers on automated document summaries and context-aware suggestions across Teams, Outlook, and Word, features that reduce the friction of knowledge work at scale. For large organizations already standardized on the Microsoft stack, the integration path is short.

Google is making a parallel push inside Workspace, using AI to automate document management and streamline workflow routing. The pitch is time savings for enterprise users across Gmail, Docs, and Drive, and for operations teams managing high-volume internal processes, the efficiency gains are meaningful even if they are harder to quantify in a CFO deck.

Salesforce's Einstein GPT targets a different function entirely: customer-facing automation. The tool allows businesses to generate personalized responses and surface operational insights from CRM data, shifting the AI value question from internal productivity to external engagement. That framing aligns more closely with where the SAP survey data says ROI is actually appearing, in customer interaction rather than headcount reduction.

ROI is real, just not where the spreadsheets said it would be

The SAP survey findings, as reported by CIO Dive, put a finer point on a tension many CIOs are navigating quietly: boards approved AI spend on efficiency grounds, but the measurable wins are showing up in analytics and engagement. That gap is not a failure of the technology, but it does require finance and IT leaders to recalibrate how they measure and report AI value internally.

Compounding that challenge is a trust problem. According to CIO Dive, a report from fintech firm Tether found that nearly half of users distrust AI companies despite widespread chatbot usage across organizations. For enterprise operators, this is not an abstract concern. Distrust among employees or customers can slow adoption, degrade the quality of AI-assisted outputs, and create compliance exposure when users route around sanctioned tools.

Governance gaps are appearing elsewhere too. CIO Dive also reported a sharp rise in AI adoption specifically for cyber defense, alongside a warning that governance frameworks have not kept pace. For CIOs managing both an AI deployment agenda and a security posture, the two pressures are colliding.

Agentic AI is the next cost pressure

Beyond the productivity and insight debate, a newer variable is starting to strain budgets. CIO Dive reported that agentic AI, systems that autonomously execute multi-step tasks rather than simply respond to prompts, is driving a significant increase in enterprise AI usage and associated costs. OpenAI, in guidance cited by CIO Dive, advised CIOs to establish clear visibility into demand, spend, and risk before scaling agentic deployments further, framing cost governance as the critical near-term discipline.

Agentic tools from all three major vendors are either available or in active rollout. Microsoft's Copilot Studio allows enterprises to build custom agents on top of the Copilot infrastructure. Salesforce's Agentforce, the company's autonomous agent platform, extends Einstein GPT into process execution. Google's Workspace is similarly building agentic capabilities into its enterprise tier. Each raises the same operational question: who owns the spend visibility when AI starts acting, not just advising.

What this means for your team

• Audit your AI ROI metrics now: if your business case was built on cost savings or time reduction and you are not seeing those returns, use the SAP survey framing to reposition value around insight generation and customer engagement before your next budget review.
• Get ahead of agentic cost exposure: map which teams are piloting or scaling agentic AI tools, establish spend visibility dashboards, and set usage thresholds before autonomous agents generate billing surprises, especially on Microsoft Copilot Studio or Salesforce Agentforce.
• Close the governance gap before it closes you: the combination of rising AI use in cyber defense, near-half user distrust rates, and accelerating agentic deployment makes a formal AI governance framework non-optional; assign ownership and document acceptable-use policies for each platform in your stack.
• Evaluate vendor AI on function-specific outcomes: Copilot, Einstein GPT, and Workspace AI serve different operational needs; benchmark each against the specific workflow it is meant to improve rather than a single enterprise-wide productivity metric.