AI ROI: Your AI Saved 10,000 Hours, What’s the IRR?

Your AI team just reported 10,000 hours saved. Great. Show me how those hours generated more free cash flow.

Did payroll fall? Did you avoid hiring? Did margins or revenue improve because of it? Did a single customer pay you sooner? If the answer is no, you may have some measure of productivity success. You haven’t proven an economic one yet.

In my last piece, I argued that AI budgets belong on broken work, not demos. Now comes the harder question: does fixing that work return enough to justify the investment? It’s an uncomfortable and growing question for CEOs and CFOs.

The return is getting harder to prove

At the end of September, BCG’s Applied AI Index reported that nearly half of companies are generating meaningful value from AI. Less than three weeks later, Gartner reported that only 13% of CIOs say their organizations have realized significant or transformational value. A separate Gartner survey found that 44% of CEOs struggle to demonstrate a financial return.

All three can be right. But one Gartner finding deserves a CFO’s attention: 58% of CIOs face growing pressure to deliver cost improvements from AI, while 51% expect AI to increase the total cost of ownership of their technology.

We’re asking AI to take cost out of the business while preparing to spend more to operate it. That may be a good investment, but it raises the bar for proving a positive internal rate of return. And we’re still measuring too much of that return in hours saved, not cash flows generated.

Hours are an input

If a salaried rep saves four hours a week and spends them in more meetings, the P&L doesn’t move. What counts is what the business does with the time.

One of the most celebrated AI wins happens in the CFO’s own office. AI can take hours, sometimes days, out of the monthly close. I’m all for that, and so is my CFO. But what changed financially? Did we cut outside costs, eliminate overtime, make better working-capital decisions or put that capacity to work on something that creates value? If not, the books just closed faster, and we’re still paying the same salaries.

Where the value leaks

In large enterprises, some of the biggest opportunities sit in transactions that never get completed correctly. That’s lost revenue and leaked margin. Consider Stanley Black & Decker. Roughly a quarter of its quotes and orders were inaccurate, contributing to an estimated $10 million a year in margin leakage. The company had already spent a year and about $1 million on a conventional integration approach that hadn’t solved the problem.

Connecting the selling process directly to live SAP pricing fixed the errors at their source. Its VP of Sales described spending a quarter of his day correcting quote errors before, and those errors disappearing once the data came straight from SAP. The win was protected margin and a sales team that stopped cleaning up its own quotes, so it could spend that time closing deals.

Legrand saved more than 50,000 hours of manual work. What mattered more was what happened next: it won 5% more bids, its people spent more time with target accounts, and Legrand estimates the combined revenue impact at $3.4 million. Revenue growth is a key value driver, but it isn’t profit or cash flow. The real return depends on the contribution that revenue generates and the cash the business collects.

Both examples came from enosix’s SAP integration work, before we built arnold® AI. They show what opens when enterprise systems execute the business correctly, and AI can accelerate it.

The Execution Gap

The market has spent years making AI better at understanding information and recommending what to do. That’s research. In a transaction-heavy business, the economic value depends on what happens next: execution. Can the order be placed at the right price, with inventory available and credit cleared, and recorded without someone stepping in? For most manufacturers and distributors, those steps live in the ERP, and for many of the largest, that’s SAP.

We call this the Execution Gap: the distance between what AI decides and what the enterprise’s systems of record will actually let it do safely and correctly.

Here’s what AI changes. Integration put the right price in front of a person, who still had to act on it one quote at a time. An AI agent can work every quote, order and exception, around the clock. But scale cuts both ways. An agent working from an approximation of the rules scales the errors, and you’re back to issuing credits and rebills. An agent working through live SAP rules scales results like the ones above.

That’s why we extended our proprietary SAP framework and built arnold® as a governed execution layer. Through SAP virtualization, AI gets secure, role-based, governed access to the live business logic already running inside SAP, so the rules don’t have to be rebuilt for every new agent. It works across ECC and S/4HANA, including mid-migration, so you’re not paying to rebuild integrations every time the AI platform or the ERP underneath it shifts.

The stakes are rising. BCG found that 42% of companies expect AI agents to have real decision-making authority by 2030, while only 5% have the full set of controls in place today. An agent that can’t execute is just advice. An agent that executes without governance is a liability that destroys value.

What I’d put in front of a CFO

There are three numbers you shouldn’t blindly accept as AI ROI: hours saved, agents deployed and pilots launched. They’re useful operating metrics. But it’s like hiring a sales rep, handing them a quota and celebrating the hire while they sell nothing. None of it proves a financial return.

Instead, every AI business case should answer five questions:

Impact: Which line of the financial statements will this change, and by how much?
Baseline: What is that number today, before we touch anything?
Cash timing: When should it move, and when will the cash actually arrive?
Investment: What will it cost to build and run, including the higher cost of ownership most CIOs now expect?
Counterfactual: What would have happened anyway, and what does doing nothing cost?

Answer those and you can hold AI to the standard you’d apply to a new plant or an acquisition: incremental free cash flow, payback, NPV and IRR. Then ask the real question: does the expected return justify the capital and risk compared with the alternatives? AI doesn’t deserve a different standard just because the technology is exciting.

A business case that can’t answer the first three isn’t ready for a capital allocation decision. One that can’t answer the last two isn’t honest yet.

To put scale on it, take a hypothetical manufacturer with $1 billion in annual revenue. If better pricing and order execution lift sales by just 0.25%, that’s $2.5 million in revenue. At a 30% contribution margin, about $750,000 a year reaches the bottom line. Put $1 million in up front and $250,000 a year to run it, and that’s roughly a two-year payback and a five-year pre-tax IRR of about 40%. It’s an illustration, not a customer result. Change any assumption and the answer moves, which is the point: now you’re debating cash flows, not hours.

Over the next few articles, we’ll take this scorecard into the governed workflows where the Execution Gap costs the most, starting with revenue execution.

Ten thousand hours saved makes a great slide. Free cash flow your CFO can find is what creates shareholder value.

About Nick Fera

Nick Fera is Chief Executive Officer of enosix and Managing Director of enosix GmbH, where he leads the company’s strategic, operational, and financial activities globally. Before enosix, he served as CEO of Firm58 and on the board of ThirdPartyTrust, which was acquired by Bitsight in 2022. Earlier, as CEO of Parlano, he led the company’s sale to Microsoft, where the technology became part of Microsoft Teams. He holds a BS in Finance from the University of Illinois and an MBA from Northwestern University.

Learn more at www.enosix.com.