There’s a question that keeps coming up in boardrooms, Slack channels, and awkward quarterly reviews: “Is our AI stuff actually working?” It’s a fair question. Companies have poured billions into AI over the past few years, and many of them are still scratching their heads about whether they’re getting real value.

The problem isn’t that AI doesn’t work. It’s that measuring its impact is genuinely hard. Traditional ROI calculations — spend X, get Y — don’t map neatly onto AI projects. So what are companies actually doing?

The Old Way Doesn’t Cut It

Most businesses are used to measuring technology investments in straightforward terms. You buy a new CRM, and you can track how many more deals your sales team closes. You upgrade your servers, and you measure uptime improvements. Simple.

AI doesn’t play by those rules. A machine learning model that improves product recommendations might boost revenue, but it also reduces customer churn, improves satisfaction scores, and gives your marketing team better data. How do you attribute all of that to one investment?

One firm we talked to pointed out that companies often make the mistake of trying to isolate AI’s impact from everything else happening in the business. That’s like trying to figure out how much of your dinner’s flavour came from the salt versus the garlic. They work together.

What Smart Companies Track Instead

The businesses getting this right tend to focus on a few key areas:

Time savings. This is the easiest win to measure. If your customer service team was spending four hours a day answering routine questions, and an AI chatbot now handles 60% of those, you’ve got concrete numbers. Hours saved, multiplied by labour costs. Done.

Error reduction. Manufacturing companies have been particularly good at this. When AI-driven quality control catches defects that human inspectors miss, you can measure the reduction in returns, warranty claims, and wasted materials.

Speed to decision. This one’s harder to quantify but arguably more valuable. When a financial analyst can run scenarios in minutes instead of days, the value isn’t just the time saved — it’s the better decisions that come from having more information sooner.

Customer experience metrics. Net Promoter Scores, customer satisfaction ratings, resolution times. These aren’t AI-specific metrics, but tracking them before and after AI implementation gives you a useful picture.

The Honest Numbers

Here’s where things get interesting. According to several industry surveys from late 2025, about 40% of companies say their AI investments have met or exceeded expectations. Another 35% say it’s too early to tell. And roughly 25% admit they haven’t seen the returns they hoped for.

That last number doesn’t mean AI failed. In many cases, it means the implementation was rushed, the problem wasn’t well-defined, or the data wasn’t clean enough to begin with. Bad inputs produce bad outputs, regardless of how sophisticated your model is.

The Hidden ROI Nobody Talks About

Something I’ve noticed in conversations with business leaders is that the biggest returns often come from places nobody expected. A logistics company implements AI to optimize delivery routes and saves on fuel costs — that was the plan. But they also discover that drivers are happier because they’re spending less time stuck in traffic. Turnover drops. Recruitment costs fall. None of that showed up in the original business case.

This is why rigid ROI frameworks can actually be counterproductive. They force you to define success narrowly before you start, which means you might miss the most important outcomes.

A More Practical Approach

If you’re trying to figure out whether your AI investment is paying off, here’s what I’d suggest:

1. Set baseline metrics before you start. You can’t measure improvement if you don’t know where you began.
2. Track both direct and indirect impacts. The second-order effects are often where the real value lives.
3. Give it time. Most AI projects need six to twelve months before they’re delivering meaningful results. Expecting a return in the first quarter is unrealistic.
4. Talk to the people using it. Dashboards and spreadsheets only tell part of the story. The humans interacting with AI tools daily have insights that data alone can’t capture.
5. Be honest about failures. Not every AI project will work. The companies that learn the most are the ones willing to admit when something isn’t delivering and pivot accordingly.

The Bottom Line

Measuring AI ROI isn’t impossible, but it requires a different mindset than most businesses are used to. The companies doing it well are the ones treating AI measurement as an ongoing process rather than a one-time calculation. They’re tracking multiple metrics, staying flexible about what success looks like, and being patient enough to let results develop.

The worst thing you can do is avoid measuring altogether. Even imperfect data is better than flying blind.