Writer’s 2026 enterprise AI survey covered 2,400 workers across global enterprises. Ninety-seven percent of executives said their organization has deployed AI agents. Only 29 percent reported meaningful return on investment. The reported rates differ by 68 percentage points, but the aggregate figures do not identify a separate group of respondents who deployed agents without returns. That requires respondent-level evidence about overlap and comparable definitions.
Separate research from the same period reports that 59 percent of companies invest more than a million dollars a year in AI. That spending measure alone says little about how the work changed or which deployments earned returns.
Review what changed in daily work
An adoption inventory records tools and access. A workflow review adds the steps people actually perform, including how they check output and pass work to the next person.
Ask whether your organization looked at a workflow, identified what an agent could own, restructured the handoffs, and measured the result. Licenses and chatbot rollouts are starting points; the test is what changed in daily operations.
The survey’s deployment and ROI rates do not establish which respondents redesigned their workflows. Use your own process record to check that connection before attributing an outcome to a particular way of deploying AI.
A reported workflow example
Microsoft’s FY26 year-in-review describes Chow Tai Fook deploying over 400 customized AI agents supporting more than 24,000 employees, with a reported 70 percent efficiency gain across core business processes. The account describes mapping agents to decision points in customer, product, and operational workflows. This is a vendor-reported example; the efficiency figure needs its original measurement definition before comparison with another organization.
For a smaller trial, identify recurring decisions and handoffs that delay completion. Give an agent a bounded part of that process, then compare the work before and after the change, including the time people spend reviewing it.
Keep a record the next trial can use
An operating record can help the next trial: it shows which tasks completed reliably, where people intervened, and what access the agent needed. Keep those details with the workflow instructions.
Organizations still searching for ROI face costs, complexity, and growing pressure from leadership to show results. That pressure can push them to add tools, licenses, and agents while leaving the quality of deployment unchanged.
Review that record before adding more agents. The survey rates do not tell us how quickly either group will learn or what its next deployment will produce.
Set a measurable workflow change
Choose a workflow where you can describe the current steps and measure completion. Specify which responsibility will move to the agent, which person will review it, and how the handoff will work.
If daily work remains unchanged after deployment, inspect where the agent’s output goes. A useful next trial should test a specific change to that path and record the result. It should also leave room for the possibility that a different tool, narrower scope, or manual process is the better choice.