ROI, agentic surge, build ceiling

Companies are starting to judge artificial intelligence less by demos and more by which stage of return on investment they can actually prove. (atlassian.com) Atlassian published a four-stage enterprise framework on March 24, 2026: exploring, optimizing, enhancing, and transforming. It argues that early projects should be measured on experimentation and time saved, while later stages shift toward team performance, new tools, and organization-wide change. (atlassian.com) Agentic artificial intelligence — software that can take actions with limited supervision instead of just answering prompts — is the next test of that model. IBM defines agentic AI as systems that pursue a goal with limited supervision, and Google Cloud said in its 2025 ROI study that 52% of executives using generative AI also had AI agents in production. (ibm.com, services.google.com) The appeal is operations work that already follows a chain: gather information, make a decision, trigger the next step. McKinsey said on April 1, 2026 that agentic systems are especially suited to procurement, manufacturing, and product-development workflows because they can automate the decision-and-action layer, not just draft text. (mckinsey.com) That is why consultants are increasingly framing AI pitches around a client’s current maturity instead of promising the same payoff to every buyer. Microsoft’s February 7, 2025 framework for agentic apps says ROI can come from cost savings, revenue increases, productivity gains, and data-quality improvements, but that the mix depends on the use case. (techcommunity.microsoft.com) The “build ceiling” sits inside that conversation. Companies with strong engineering teams often prefer to build custom systems up to a certain complexity or strategic importance, then buy outside tools only when speed, integration, or maintenance costs outweigh the control of doing it in-house. (atlassian.com, techcommunity.microsoft.com) The numbers behind the broader shift are still uneven across studies. Google Cloud said 88% of agentic AI early adopters reported positive return on investment on at least one generative AI use case, while Capgemini said in June 2025 that organizations surveyed were seeing an average 1.7 times return from their artificial intelligence and generative AI investments. (services.google.com, capgemini.com) Those gains are not universal. Capgemini said one in five organizations already used AI agents or multi-agent systems in 2025, but McKinsey said most operations companies were still at an early stage and cited research consensus that about 90% of companies attempting these transformations do not see real financial benefit. (capgemini.com, mckinsey.com) Microsoft’s workplace research points to another constraint: readiness inside the company. In a study of 1,800 full-time employees across nine markets conducted from June 11 to July 7, 2025, Microsoft said only one in six leaders reported that their organization had fully established transformation strategies for agentic AI. (techcommunity.microsoft.com) So the practical question in 2026 is narrower than “should we use AI.” It is whether a company can show savings in the first stage, expand into revenue or workflow redesign in the next one, and know exactly where its own build ceiling ends before the spending starts. (atlassian.com, techcommunity.microsoft.com)