Most companies buying AI agents externally report either adoption without realized returns or returns limited to pilots, with only 12.6% achieving sustained ROI at scale. This gap between AI deployment and actual business impact reflects a deeper problem: enterprises are automating work without the oversight, process knowledge, and employee adoption infrastructure required to turn automation into measurable value.

Why Are Companies Struggling to Measure AI Agent Returns?

The challenge isn't technology availability. It's governance and accountability. When AI agents operate across business processes without clear ownership, risk assessment, or connection to company rules, the results become unpredictable. A recent survey of 833 enterprise software decision makers found that expected and actual returns on recent software purchases both average roughly 14 percent, yet 41.5 percent of decision makers say validated ROI studies would increase their confidence in allocating budget for automation investments.

This confidence gap matters because companies are making large bets on AI agents without clear benchmarks for success. The problem isn't that automation doesn't work; it's that enterprises lack the frameworks to connect automated actions with measurable business outcomes. Without process intelligence, governance oversight, and employee adoption tracking, companies can't tell whether an AI agent is genuinely improving performance or simply shifting work around.

What Does Effective AI Agent Governance Look Like?

SAP's recent expansion of its Business Transformation Management portfolio offers a concrete example of how enterprises can address this gap. The company announced updates to its Signavio, LeanIX, WalkMe, and Cloud ALM tools at its Transformation Excellence Summit on September 22, 2026, designed to help organizations deploy, govern, and monitor AI agents in production environments.

The approach centers on three critical elements: connecting AI agents to company knowledge and process rules, establishing clear accountability and risk assessment, and ensuring employees actually use the tools consistently in their daily work. Without these elements, automation remains isolated and unmeasured.

Steps to Build AI Agent Governance Into Your Enterprise

• Map Process Knowledge: Use process mining to identify where AI agents can help, then connect agents to a shared repository of company rules, operating constraints, and decision-making logic so agents operate within established guardrails.
• Establish Clear Ownership and Risk Assessment: Assign accountability for each AI agent, automate risk classification using compliance frameworks like the EU AI Act and NIST standards, and reassess risk when systems change to ensure agents remain aligned with regulatory requirements.
• Track Employee Adoption and Workflow Changes: Monitor whether employees actually use AI assistance consistently in their daily work, measure cycle-time and cost improvements with clear baselines, and use workflow analytics to identify changes in working practices over time.
• Measure Outcomes in Production: Deploy agents with analysis, configuration, testing, and rollout processes, then track agent actions in production to identify deviations from intended behavior and assess whether automation is delivering promised returns.

What Real-World AI Agent Improvements Look Like

Early examples suggest where AI agents can deliver measurable value when governance is in place. A cement manufacturer deployed a voice-based agent to capture purchase requisitions and create purchase orders after approval, reducing a cycle previously measured in weeks. A conglomerate handling more than 15,000 requests for proposals annually used Joule Agents (AI assistants built into SAP's platform) to speed responses. A consulting firm saw evaluation work that previously took six hours completed in seconds with AI assistance.

However, these examples lack independent verification, and implementation still requires business sponsorship, employee participation, and reliable data. The lesson is clear: enterprises should measure their own cycle-time and cost improvements, including token costs for running AI models, and require pilots to demonstrate a credible path to production before scaling.

Why Employee Adoption Remains the Hidden Bottleneck

The gap between access and successful use of AI remains wide across enterprises. A September 2026 survey of chief information officers found that 73.3 percent of organizations buying AI externally report either adoption without realized ROI or ROI limited to pilots, and only 12.6 percent report sustained ROI at scale. This suggests that the problem isn't whether AI agents work in theory; it's whether employees use them consistently in practice.

WalkMe's interface-based execution tools address this adoption challenge by bringing AI agents into existing workflows across legacy systems, customized enterprise resource planning environments, and third-party applications that standard API connections cannot reach. The practical measure of adoption is whether employees use these capabilities consistently in their daily work, not whether the technology exists.

What Should Enterprises Expect From AI Agent Investments?

The data suggests realistic expectations matter. When enterprises invest in AI agents without governance, process knowledge, and adoption infrastructure, they often see pilots that work but don't scale. When they invest in the full governance stack, they can measure improvements in procure-to-pay cycles, accounts payable processing, and receivables management, including metrics like days sales outstanding.

The future of enterprise AI success depends less on the sophistication of the AI agent itself and more on whether companies can connect automation to their own processes, establish clear accountability, and ensure employees actually use the tools. Without these elements, even the most advanced AI agents become expensive pilots that never reach production scale.