Neda Nia drives Stibo Systems' product vision, shaping strategy, innovation, and growth to create measurable value for customers.

getty

​Most companies still can't answer a basic question: Is our AI investment actually working?

It's not for lack of attention, but because they're using the wrong metrics.

For the past couple of years, I've watched enterprises pour serious money into pilots and initiatives and then struggle to explain what came back. To be fair, the ask from the boards was simply to adopt AI. While CIOs and CFOs were nervous to adopt AI, the priority was to adopt it and measure.

The problem is that we keep trying to measure something fundamentally new with metrics designed for legacy operations.

The demo worked, but the business didn't change.

There's a pattern we keep seeing: A pilot gets approved, the results look promising and someone screenshots a productivity win, then not much else happens.

Most enterprises started by layering these tools on top of existing workflows. It felt like the low-risk move: run a test, see what sticks, then scale. But that approach has a hidden cost. You end up paying for technology that accelerates a broken process. You just have a faster version of the same problem. And when the workflow itself doesn’t fundamentally change, leadership struggles to point to meaningful business impact, even when the technology looks successful on paper.

That gap between the impressive demo and real business change is the cost that most finance teams won’t be able to measure. Hence, the ROI can’t be identified.

You can't just layer it on.

The instinct in most organizations has been to drop new tools on top of existing workflows and hope for efficiency gains. That's a bit like hiring a brilliant new colleague and then giving them no context about how decisions get made, no access to the right information and no authority to act on anything.

It doesn't work.

Real value shows up when organizations redesign the workflow itself, not just what sits on top of it. That means flattening the information bottlenecks that slow every team down, reducing the coordination back-and-forth that consumes hours each week and shifting people from being the ones who execute every step to being the ones who guide, review and decide.

Often, the hardest part is the human side, not the technology. Old habits and entrenched hierarchies don't shift because you added a new tool. Jack Dorsey's framing resonates here: Rather than moving from human to machine, the real move is from hierarchy to intelligence. That means amplifying human agency and judgment, not automating around the existing structure.

The ROI question is really a workflow question.

CFOs are asking the right thing: Where is the return? But the math is harder than it looks.

The economics here are genuinely unusual. Infrastructure costs shift as you scale. Not to mention, the LLM providers and the cost of running these systems fluctuate in ways that don't map neatly to familiar budget categories. Productivity gains are real but hard to isolate from everything else happening at once. And the "experimentation" phase, which many companies are still in, is expensive by design.

Here's what the fastest-learning organizations are discovering: It exposes friction that was always there and costs that people had been absorbing. That hidden cost is now visible, which is actually an opportunity, if leadership is willing to do something about it.

Trust is the ceiling.

Here's where it gets interesting. As these tools take on more of the actual work, not just surfacing information but helping make real decisions, trust becomes the thing that determines how far they can go.

This is practical trust: Can we stand behind this output? Is it grounded in our actual data? Does it reflect our specific policies and context?

Organizations that invest in clean, well-governed, reliable information are going beyond good housekeeping and building the foundation that determines how useful any of this can actually be. The better the underlying infrastructure, the more confidently people can act on what the system produces, and the more of the actual work it can take on over time.

This is where under-investment can get expensive. Fragmented data, outdated systems and unclear governance put a ceiling on what's possible.

Consider the open question.

One more thing worth watching is who owns the intelligence.

Companies building their capabilities inside closed, proprietary systems are betting that one vendor will have all the answers, indefinitely. That's a risky bet. Vendor lock-in in this space can mean losing control of your own business context: the logic, the data and the way your organization actually operates. The cost of switching later, financially and operationally, can be significant.

The smartest enterprises are thinking carefully about flexibility from the start: owning their own business context, keeping their options open and making sure the convenience of today doesn't become the dependency of tomorrow.

What's the bottom line?

The real cost of AI adoption isn't the software license or the compute bill. It's the organizational work: the redesign, the reskilling, the governance and the change management. That work is what turns a promising tool into a different way of operating.

Most enterprises are budgeting for the tool. The ones pulling ahead are budgeting for the transformation.

The enterprises that win in this next chapter will be the ones that redesigned how intelligence moves through their organization and invested in the people and process changes to make that actually stick. That's not a technology question. It's a leadership one.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?