How to Measure AI ROI (Without Kidding Yourself)

I've talked to over 500 businesses about AI in the last year. The question I hear more than any other is not "which tool should we use?" It's "how do we know if this is even working?"

That's the right question. And almost nobody has a good answer.

A Harness survey from 2026 found that 26% of enterprise AI spending is wasted. That's one in four dollars going nowhere. And 56% of organisations describe their AI cost forecasting as "guesswork."

It gets worse. Flexera's 2026 ITAM report found that 59% of organisations say their AI waste actually increased over the past year. Only 31% have accurate visibility into what they're even spending on AI.

Let that land. Most companies are spending more, wasting more, and measuring less.

I built an AI company that tracked client retention metrics down to the individual appointment level. We identified over $650,000 in missed revenue for a single client in one quarter. Not because the AI was magic. Because we measured the right things from day one.

That's what this article is about. Not a finance textbook. Not a dashboard template. A real way to figure out if AI is pulling its weight in your business.

Why is it so hard to measure AI ROI?

It's hard because AI touches everything and owns nothing. A chatbot handles customer questions. Did that reduce support tickets? Maybe. Did it reduce churn? Unclear. Did it annoy three of your best clients so badly they left? Nobody checked.

Traditional software ROI is simpler. You bought a CRM. You track deals. You see revenue. Done.

AI is different for three reasons.

First, AI often improves a process instead of replacing it. That makes the impact harder to isolate. Your team still does the work. AI just made it faster, or caught something they missed. How do you put a number on "caught something they missed?"

Second, most businesses measure AI activity instead of AI outcomes. They track how many times the tool was used. How many queries it answered. How many documents it summarised. None of that tells you if the business got better.

Third, and this is the big one, you can't measure AI ROI if your systems were broken before AI showed up. If your data is messy, your workflows are unclear, and your team doesn't know who owns what, then AI just automates the mess faster. The ROI you're measuring is actually the cost of your broken foundation.

If your systems are broken, AI just automates the mess faster. Fix the foundation first.

What should you actually measure?

Forget "time saved." Start with business outcomes that existed before AI arrived. Revenue. Retention. Error rates. Response times. Cost per unit of work. Client satisfaction scores.

The question is not "what did AI do?" The question is "did this number move after AI got involved?"

Here's a simple framework I use with every business I work with. Three layers. No spreadsheet required.

Layer 1: Did the outcome change?

Pick 2 or 3 business metrics that AI is supposed to improve. Not AI metrics. Business metrics.

For example, when I built Latch AI, we measured client retention rates and recovered appointments. Not "AI engagement" or "model accuracy." Retention and appointments. Those are numbers the business already cared about.

We grew retention 3.6% across an entire book of business in 90 days. We recovered 77 extra appointments in the first 8 weeks that were falling through the cracks. Those are outcomes. They matter because the business already tracked them. If retention is your number too, here's how AI catches clients before they leave.

If your AI tool doesn't connect to a business metric the CEO already watches, you have a hobby project. Not an AI initiative.

Layer 2: What would have happened without it?

This is where most people stop measuring and start guessing. You need a baseline. What was happening before AI?

The simplest version: compare the same metric from the same period last year, or from the 90 days before you turned AI on. Account for seasonality and obvious changes (new hires, lost clients, a product launch).

You don't need a data science team for this. You need a before number and an after number. And you need to be honest about what else changed in between.

If retention was 82% before AI and 85% after, but you also hired two new account managers in the same window, AI didn't do all of that. Maybe it did some. Be honest about the split.

Layer 3: What did it cost?

Not just the subscription. The real cost.

Add up the tool subscription, the hours your team spent setting it up, the hours spent maintaining it, the training time, and any integration costs. That's your actual investment.

Then compare it to the outcome change from Layer 1.

If AI helped you recover $50,000 in revenue you were losing and it cost you $12,000 all in, that's a number you can take to your team. If AI saved your admin person 3 hours a week and it costs $200 a month, you can decide if that's worth it.

The math does not have to be complicated. It has to be honest.

How do you know if AI is worth it for your business?

Here's the honest answer. AI is worth it when it changes a business outcome more than it costs, and your team actually uses it.

That second part is the one everybody skips. McKinsey's 2025 Global AI Survey found that 88% of organisations now use AI in at least one function. But the majority are still stuck in pilot mode. Only 7% have fully scaled AI across their operations.

The gap between "we have AI" and "AI is working" is where most businesses live. I see it every week. A company buys a tool, runs a pilot, sees a promising demo, and then nothing changes. The team goes back to the old way. The tool collects dust. And six months later, someone asks "what happened with that AI thing?"

What happened is nobody defined what "working" meant before they started.

Before you buy a single AI tool, answer these three questions:

1. What specific business outcome should improve? Not "efficiency." Something you can measure with a number that already exists.
2. What's the baseline right now? Measure it for at least 30 days before AI touches it.
3. What would make this worth the investment? Set a threshold. If retention improves by 2%, it's worth it. If it improves by 0.3%, it's not. Decide that before you see the results.

If you can't answer those three questions, you're not ready for AI. You're ready for an AI operations audit to figure out where AI actually fits.

What are the biggest AI spending mistakes businesses make?

After 500+ conversations with businesses about AI, I see the same mistakes over and over.

Measuring activity, not outcomes. "The tool processed 10,000 documents this month." Great. Did anything change? Did errors go down? Did the team move faster? Did a client notice? If you can't connect the activity to an outcome, the activity is meaningless.

No baseline. You can't prove AI improved something if you don't know where you started. This is the most common mistake I see. A business turns on AI and says "it feels faster." Feelings are not ROI.

Too many tools, no integration. The Harness survey found that 52% of organisations have no clear owner for AI costs. Tools get bought by different departments. Nobody connects them. Nobody measures the overlap. You end up paying for three tools that do the same thing and a fourth that nobody uses.

Confusing a pilot with proof. A pilot shows you that something can work. It doesn't show you that it will work at scale, in your business, with your team. The gap between pilot and production is where most AI investments go to die.

Ignoring what AI breaks. AI can speed up one process while quietly damaging another. An AI chatbot might resolve tickets faster. But if it gives wrong answers 8% of the time, you just created a new problem while solving an old one. Measure the side effects, not just the intended effect.

A real way to track AI ROI in 90 days

Here's the process I recommend for any business that wants a real answer. Not a dashboard. Not a report for the board. A genuine understanding of whether AI is pulling its weight.

1. Pick one workflow. Not five. One. The one that costs the most, takes the longest, or loses the most money. If you don't know where to start, read why most AI tools fail before you try to fix everything at once.

2. Measure the baseline for 30 days. Track the business metric attached to that workflow. How long does it take? How much does it cost? What's the error rate? What's the output? Write the number down.

3. Turn AI on and measure for 60 days. Same metric. Same conditions as close as possible. Track it weekly so you can see the trend, not just the average.

4. Do the math. Compare the before and after. Subtract the cost of the AI tool and the time to implement. If the number is positive and meaningful, keep going. If it's flat or negative, either the tool is wrong, the workflow isn't ready, or the foundation needs fixing first.

This is exactly how we worked at Latch AI. We picked client retention. We measured the baseline. We turned the AI on. And we tracked outcomes week over week. When we found nearly $500,000 in recoverable revenue for another client, that wasn't a guess. It was the math.

You can do the same thing. You don't need a data team. You need a baseline, a business metric, and 90 days of honesty.

What should your next step be?

If you read this and realised you don't know whether your AI tools are working, that's not a failure. That's where most businesses are right now. The failure would be ignoring it.

Start with one workflow. Measure the baseline. Give it 90 days. And be willing to hear the answer, even if the answer is "this tool isn't doing what we thought."

If you want someone to do this with you, that's what the AI Business Audit is for. Two weeks. Real numbers. A plan you can act on. If the gap is that your team never got aligned on AI in the first place, an AI workshop starts there instead.

Either way, book a call and tell me what's going on.