We're Asking the Wrong ROI Question About AI

By Alexander Hagerup, Co-Founder & CEO, Vic.ai

The question I get most from finance leaders is a simple one: how do we calculate the ROI of AI?

I've been thinking about that question for almost a decade now. When Kris and I started Vic.ai in 2017, we spent the first two years working only on the AI itself, training on hundreds of millions of accounting transactions before we built a single screen of the product. So I have a lot of respect for CFOs who want proof before they invest. It's the right instinct.

But I've come to believe we're asking the ROI question the wrong way.

For years, we've evaluated AI the same way we've evaluated every enterprise technology before it. Labor savings, process efficiencies, implementation costs, payback periods. All of that matters, and it belongs in any business case (we hold ourselves to those numbers with every customer). But it doesn't capture what I believe will be the defining advantage of AI over the next decade.

The organizations pulling ahead won't just be the ones that deploy AI. They'll be the ones that reorganize how they operate around what AI makes possible. And that's a very different conversation.
‍

The returns that matter most are the hardest to put on a spreadsheet

In finance, it's easy to reduce this discussion to invoices processed, hours saved, headcount avoided. Those outcomes are real, and you should absolutely expect them from any technology you buy.

But here's what I hear from the finance leaders who have been running on AI the longest: the returns they care about most show up somewhere else. Better decisions, because information arrives sooner. Stronger working capital, because the team can actually see what's happening with spend. Faster closes. Fewer surprises. A finance organization that spends its time guiding the business instead of documenting what already happened.

And unlike labor savings, which you capture once, these advantages compound. They get more valuable every year you have them.
‍

Build versus buy is a bigger comparison than it looks

A few months ago, one of our largest customers emailed me asking whether, given how accessible generative AI has become, they should just build this capability themselves. I wrote a whole piece about that question because I think it's the most interesting conversation happening in enterprise software right now. At the same time, I think most teams underestimate what they’re actually comparing. 

Building an enterprise AI application isn't building a model that does a task. It's building a production system that stays accurate, secure, and governed while sitting inside the most business-critical workflows you have. I know what that takes because we've lived it: years of iteration, millions of transactions, thousands of edge cases you'd never predict on a whiteboard. The demo is easy. The production system is not. [In fact, I wrote an article about this exact problem statement several months ago].‍

I sometimes compare it to a self-driving car. You don't want a self-driving car that's pretty good. You want the one that doesn't crash. And the best AI keeps learning from your business, which means the gap between good and great widens over time.

So the real comparison isn't software licensing versus engineering salaries. It's opportunity cost, operational risk, and time.
‍

Every major shift changes the basis of competition

I've been building technology companies since before the cloud era, and I've watched this pattern play out more than once. Cloud changed how companies compete. Mobile did. The internet certainly did. AI won't be any different.

And the companies that benefit most won't necessarily be the ones spending the most money or grabbing the newest models first. They'll be the ones that learn to weave AI into how they actually operate, faster than everyone else. That learning takes time. You can't buy it later.
‍

A final thought

Maybe that's the ROI question we should be asking. Not "what return will AI generate?" but "what does it cost us to wait while everyone else learns faster?"

The leaders I talk to who made this shift early all say a version of the same thing: the efficiency gains got the project approved, but the way their team operates now is the part they'd never give back.

Related reading: The AI ROI Question is Coming

‍