What is the Return on your Agent?

There’s a question that should be on every boardroom agenda right now, but isn’t. Not “are we deploying AI?” Not “how many use cases do we have?” The question is simpler, and most rooms aren’t ready for it: what is the return on your agent?

The conversation around scaling AI has been dominated by the technical side. Connect your systems. Clean your data. Build governance. All of that matters. None of it is wrong. But it is incomplete.

We’ve been here before. SaaS promised to transform businesses and it did. But not before someone asked what they were actually paying for.

In the early days of SaaS, businesses rushed to adopt cloud software, signed contracts, and onboarded teams. Then someone in finance squinted at the invoices and asked: what are we actually getting for this? That question changed how the industry thought about software spend. Renewal rates, seat utilisation, cost per outcome. It matured everything.

We’re at that same point with AI. But the economics are harder. With SaaS, costs were predictable: a fixed seat fee, a known renewal. With AI agents, you pay by consumption. Every query, every task, every automated decision has a cost attached to it. Run agents across thousands of workflows and those costs add up fast, and finance only notices when the invoice lands.

Deploying your first agent is rarely cheap. You are not just building the agent, you are building what it needs to function: connecting systems that have never talked to each other, bringing data together, creating the plumbing that makes automation possible. That is real investment, and it should be treated as such.

But here is where it gets interesting. That foundation, once built, doesn’t need to be built again. The second agent uses it. So does the fifth and the tenth. The cost of each new agent drops because the hard work is already done. And something else happens too: when you unify your data to make agents work, you often unlock value that was always there but never visible. Done well, the economics of AI improve as you scale in more ways than one. The challenge is that most organisations never get there, because they deploy agents without a clear business case and run out of confidence before they reach the point where it starts to pay.

That is why business case has to come first, every time. Not as a box-ticking exercise, but as a genuine test. Does this agent solve a real problem? Does it save meaningful time, reduce a real cost, or create something that wasn’t possible before? If the answer isn’t clear before deployment, it won’t become clear after.

That is the shift from ROI to ROA. Return on Agent. Define what it must return before you deploy it. Time saved. Revenue generated. Errors eliminated. And what it costs to run, at volume, over time. Not just at launch.

Winning with AI is not about having the most agents. It is about knowing which ones are actually working, what they cost to run, and building in a way that makes the next one easier to justify than the last 

That discipline doesn’t have a name yet. It should.  So before your next AI investment is approved, ask the one question that changes everything: what is the return on this agent?  If no one in the room can answer it, you have your answer.