What's the ROI of an AI agent, and how do you actually measure it?

A VP of Operations at a logistics company told her board that the company's new AI agent had saved 4,000 hours in its first quarter. The number came from multiplying ticket volume by an average handle time the team had used for years, before the agent existed.

Nobody had checked whether the agent created new work elsewhere. Two months later, her team found that support staff were spending hours a week correcting the agent's outputs before sending them to customers. The 4,000 hours were real. So was the new work. The board had only heard about one of them.

This is not an unusual story. It's close to the default outcome when a company treats ROI from AI as a number to report rather than a system to build.

Which AI agent has the best ROI?

None, as a fixed property of the vendor. Type that question into any AI search tool right now and you'll get a list of vendors: Salesforce, Microsoft, ServiceNow, Google, UiPath. That answer treats ROI like a feature the vendor ships. It isn't. ROI is what happens after deployment, and it depends entirely on what an organization decides to track, not on which logo is on the product.

A Gartner survey of 782 infrastructure and operations managers, published in April 2026, found that only 28% of AI use cases in IT infrastructure and operations fully succeed and deliver ROI, while one in five, 20%, fail outright. The gap between those numbers isn't mostly a technology problem. It's a measurement problem. Most of the enterprises in that survey can tell you an agent is running. Fewer can tell you what it's actually worth.

How do you calculate ROI on an AI agent?

By measuring three things together, not one: time saved, revenue influenced, and AI proficiency. A single figure, like hours saved, is not ROI on its own.

Time saved is the number everyone reports first, and it's the easiest one to get wrong. Time saved only means something if it's measured net of new work the agent introduces, corrections, escalations, review cycles. Reporting the hours removed without checking the hours added is how a real deployment ends up with a fictional ROI number.

Revenue influenced matters for agents that touch revenue directly: a sales assistant that shapes deal velocity, a support agent that affects renewal conversations. This requires connecting conversation-level activity to downstream business outcomes, not assuming that usage volume translates to revenue on its own.

AI proficiency gets skipped most often, and it's not a return metric by itself. It's the lever that determines whether the other two numbers improve or stall. An organization where employees ask an agent good questions and use its answers well will see better time-saved and revenue-influenced numbers than one where the same agent sits underused. Proficiency is what you improve. Time saved and revenue influenced are what you report.

Why do AI agent ROI numbers often turn out to be wrong?

Because they usually count only one side of a ledger. The pattern in the logistics example repeats across almost every AI ROI claim that later gets walked back. A number gets reported that only counts the work removed, not the work added, the corrections, the rephrased queries, the conversations that ended in a human taking over. That side of the ledger doesn't show up in a deployment dashboard. It shows up in the conversations themselves.

That's the part missing from most ROI conversations happening right now, and it's the part that determines whether the number a board hears next quarter is closer to the truth than the one they heard this quarter.

Nebuly

Getting a full picture of AI agent ROI means seeing both sides of that ledger at the conversation level, not just the summary numbers a deployment dashboard reports. That means tracking what an agent actually resolves, what it hands back to a human, and how usage patterns change as an organization gets better at working with it, across every conversation, not a sample. That's the layer most ROI conversations are currently missing.

Nebuly is the ROI platform for enterprise AI. It connects to the AI agents your business runs on, the assistants your customers interact with, and the tools your employees use every day, including Claude, ChatGPT, and Copilot, and translates that activity into business value. How much time is being saved across teams. What revenue your AI is influencing. What adoption and AI proficiency look like in practice, across departments and geographies. All aggregated at the organizational level, never tied to individuals.

If you need clarity on what your AI investment is actually delivering, book a demo.