On June 9, 2025, the IBM Institute for Business Value published “From AI projects to profits,” drawing on two executive surveys conducted with Oxford Economics to discuss how companies can move from short-term AI pilots to sustainable financial results. The research does not support two quantitative claims in the original draft, so this article instead uses the verifiable pilot and at-scale ROI figures.
IBM notes that generative AI pilots in 2023 reported early ROI of about 31%, but after scaling, the average fell back to about 7%; looking back over the past three years, surveyed CEOs said only 25% of AI initiatives delivered the expected ROI. The top 10% of organizations reached about 18%, showing that high returns are not impossible, but pilot results cannot be directly extrapolated into a guarantee at scale.
The value of Agentic AI does not lie in unconditionally handing entire processes over to “digital employees.” Companies must still define what data agents can read and write, what actions they can take, human approval points, accountability for errors, and recovery methods. Without trusted data, cross-system integration, and clear decision rights, even the most powerful model may only add cost and operational risk.
From the perspective of “The CIO for Your Business Operations,” Unima recommends establishing four elements of outcome governance: first, select a single process and record baselines for cost, cycle time, quality, or revenue; second, restrict data and system permissions and retain human approval for high-risk actions; third, track accuracy, exception rate, cost per task, and manual rework in parallel; fourth, define scale-up, stop, and rollback conditions in advance. This is an implementation checklist compiled from IBM’s research and does not represent specific client results or a return-on-investment commitment.
Official source: IBM Institute for Business Value, “From AI projects to profits: How agentic AI can sustain financial returns,” 2025-06-09, https://www.ibm.com/thought-leadership/institute-business-value/report/agentic-ai-profits. This article was verified on 2026-08-24; survey results reflect the respondent sample and its reporting basis and do not mean every company or AI agent project will achieve the same results.