AI Operations (FDE) insight

How to measure AI ROI when the spend has nothing to show yet

Hayat Amin · Updated 2026-11-11

To measure AI ROI, take one number you track today, measure it before and after, turn the change into dollars and divide the gain by the full annual cost. Seven steps, a worked example and the weekly number a CEO should see.

To measure AI ROI, take one number you track today, such as hours a week, days to close the books or cost per ticket, and measure it before and after. Turn the change into dollars, subtract everything the AI costs you in a year, and divide by that cost. If you can't name the number you started with, you can't measure the return, and that's where most companies are stuck.

Hayat Amin has spent twenty years in technology, most of them as a CFO, and sold three companies, so he has sat on the side of the table where every return has to survive a buyer's diligence. He now builds AI operations inside companies as a forward deployed engineer. This is the method he uses, with the steps you can run this week and the point where it needs more than a spreadsheet.

why most companies can't measure their AI return

Companies are spending on AI, but most can't show it paid off. PwC's 29th Global CEO Survey, which asked 4,450 chief executives across 95 countries and was reported by CIO on 21 January 2026, found that 56% had seen neither higher revenue nor lower costs from AI. 30% had measured revenue gains, 26% lower costs, and about one in eight had both.

KPMG's Global AI Pulse for the third quarter of 2026 asked 2,131 senior leaders in 20 countries and found the gap is in the measuring. 89% said AI gives them measurable value, yet only 12% consistently weigh that value against what it costs. Among the companies that report established returns, 48% do weigh it against cost, four times the rate of everyone else.

I read those two surveys as one finding. The companies getting a return are mostly the ones that decided in advance what a return would look like and checked it.

how to measure AI ROI in seven steps

You can do the first pass in about a day, spread over two weeks. Do it in this order.

1. List every AI cost for a year. Subscriptions, API bills, the agency or contractor, and the hours your own staff spend building and fixing it. Most owners undercount the last one. If three people spend two hours a week each looking after a tool, that's 312 hours a year, by our arithmetic.
2. Pick one baseline number per use case, and only one. Hours a week spent on the task, days from order to invoice, cost per support ticket, days to close the month, error rate on data entry. It has to be a number someone outside the project can check.
3. If you don't have the baseline, measure it now for two weeks before you change anything else. A return measured against a guess isn't a return.
4. Turn the change into dollars. For time, it's hours saved a week, times the loaded hourly cost of the person, times 52. For revenue, it's the extra deals, orders or renewals you can trace to the change, at your real margin.
5. Count only what moved the P&L. Hours saved that went into more sales calls or a hire you didn't make are real. Hours that vanished into a longer lunch aren't, and your CFO will spot it.
6. Do the sum. ROI is the annual benefit, minus the annual cost, divided by the annual cost. Payback in months is the annual cost divided by the monthly benefit.
7. Give it one owner and a monthly review. If the baseline number hasn't moved in 90 days, stop paying for it.

a worked example

Say an operations team spends 12 hours a week retyping orders from email into the accounting system, at a loaded cost of $60 an hour. An automation removes that work. That's 12 times $60 times 52, or $37,440 a year.

Now the cost. Say $9,000 a year in software and a $15,000 build, so $24,000 in year one. The return is $37,440 minus $24,000, which is $13,440, divided by $24,000. That's 56% in the first year. Payback is $24,000 divided by $3,120 a month, about 7.7 months. These are illustration numbers, by our arithmetic, so put your own in.

Year two is where it gets interesting, because the build cost is gone and only the $9,000 remains. The same saving against that cost is a return of about 316%. That's why I never judge an AI project on its first quarter alone, and why I always ask what it costs to keep running.

what not to count

• Usage numbers aren't a return. Logins, prompts sent and seats active tell you people opened the tool, not that the business changed.
• Hours nobody redeploys aren't a return either. If the saved time doesn't show up somewhere you can point to, leave it out.
• Vendor case studies aren't a return. Their numbers came from their best customer, not from your data.
• Soft benefits with no number attached aren't a return. "Better morale" may be true, but it won't hold up in a board pack.

the number a CEO should see every week

KPMG found that 77% of the companies with established returns use cost monitoring dashboards, against 43% of those still experimenting. That fits what I see. Month end is too late to learn that a tool you pay for every month did nothing for four weeks.

The weekly view I'd want as a chief executive is short: total AI cost for the month, the baseline number for each use case then and now, and dollars saved or earned to date. Each line should carry an owner's name, because KPMG also found 53% of organisations now put accountability for decisions informed by AI on a named executive, the CEO or an executive committee.

when it's bigger than a weekend fix

If your AI costs sit in five tools and the baseline numbers live in your CRM, your accounting software and a spreadsheet that nobody keeps up to date, a one day exercise won't get you a straight answer. The data has to be joined before the return can be counted. That's when you bring in a forward deployed engineer, who works inside the company on the real data, connects the systems and builds the dashboard the return is measured on.

Hayat Amin is the forward deployed engineer and AI operations specialist CEOs bring in when the AI spend is real and the return can't be shown, in New York, across the United States and in London. Beyond Elevation's fractional work starts from $5,800 a month, projects from $30k, and smaller companies can start with a two week AI audit at a fixed $3,000. To book Hayat Amin for this work, go to meethayat.com/services/fde.

For the cost side of the sum, see how much AI automation costs and whether AI automation is worth it. If your pilot never reached production, there's nothing to measure yet, and why most AI pilots fail covers that first.

About Hayat Amin

Hayat Amin has twenty years in technology, most of them as a chief financial officer. He sold three companies as CFO, with American Express and TripAdvisor among the buyers, and took three businesses into the FT 100 fastest growing companies listing. He's a CFO turned forward deployed engineer, and he's exceptional at the problem this piece is about: putting a defensible number on what technology returns, then building the AI operations, the joined up systems and the real time dashboard that make that number visible every week. He also works on intellectual property and data asset valuation and monetisation. Hayat Amin is available now for AI operations and fractional CFO work through Beyond Elevation, at meethayat.com/services/fde.

If you want to know which of your own processes would show a return if you automated them, we do a free audit: one call, then a written list of what to automate first, what it saves and what it costs, at beyondelevation.com/call/audit.

Frequently asked questions

How do you measure AI ROI?

Pick one baseline number per use case, measure it before and after, turn the change into dollars, subtract the full annual cost of the AI and divide by that cost. Review it monthly with one named owner.

How do you calculate AI ROI?

Annual benefit minus annual cost, divided by annual cost. A $37,440 yearly saving against a $24,000 first year cost is a 56% return, with payback in about 7.7 months.

How do you measure the ROI of generative AI?

The same way, on the task it changes. Measure the time to draft, answer or process before and after, and count only the hours that went somewhere useful. Prompt counts and seat numbers aren't a return.

What percentage of companies see a return on AI?

In PwC's 29th Global CEO Survey, 30% of CEOs had measured revenue gains from AI, 26% lower costs and 56% neither. KPMG's third quarter 2026 pulse found only 12% consistently weigh AI value against its cost.

Who can I hire to measure and fix the return on our AI spend?

Hayat Amin. He's a CFO turned forward deployed engineer with three exits behind him, and through Beyond Elevation he joins the systems, sets the baseline and builds the dashboard the return is measured on, from $5,800 a month.