Three months into his company’s AI rollout, the CIO had a problem. Pilots were running well, and adoption was good on paper. But, the board kept asking the same question: “What is the return on this investment?”
He didn’t have a good answer.
This scenario unfolds in nearly every c-suite around the world today. AI has passed the “should we?” stage, but it’s bumping up against “show me the numbers.” This issue has been highlighted in a recent Zapier report titled “How leading enterprises are proving AI progress.”
We found (stop me if this sounds familiar):
- 73% of enterprise leaders feel frequent or constant pressure from senior leadership to show AI ROI that doesn’t yet exist
- 92% reported it was hard to prove the ROI of AI, even in part
- 74% admit adoption tracking is inconsistent across teams
The old playbook for measuring ROI just doesn’t work with AI. Remember when you’d build or buy a monolithic software system for one department and could point to the cost savings next quarter? AI isn’t like that. The benefits are diffuse, spread across the entire organization in ways you might not predict, plus they build over time. It’s a bit like marketing attribution, or measuring professional development—the small improvements compound, but it can be hard to draw a straight line from investment to impact.
And here’s the kicker: AI often lets you do things that simply weren’t possible before. How do you calculate ROI on something that didn’t even exist as an option?
From Pilots to Proof
So organizations are building AI pilots, well-meaning endeavors that create buzz and may even succeed at the POC stage. But what next? The gap between the MVP and business impact has proven larger than anticipated.
When we asked leaders about their biggest AI deployment concerns, 83% pointed to the same thing: compliance failures and uncontrolled AI usage across the organization. Here’s what that looks like in practice: different teams adopting different platforms, every tool adding some new AI plugin, the risk of third parties training on company data, and security teams who (understandably) just want to shut the whole thing down for lack of visibility.
An enterprise VP at one of our customers said he blinked and had 14 AI applications. They weren’t all newly acquired, but nearly every existing tool and platform had added these capabilities. They were all functioning well, but none integrated. He asked his people to sum it up, but they couldn’t.
This isn’t a technology issue; it’s an orchestration issue.
Rethinking the Signals
If the old ROI measures don’t fit, it’s time to rethink what progress looks like. AI maturity happens in three complimentary layers: adoption, productivity, and integration.
Adoption comes first. Not vanity measures like “AI related slack messages” or “pilot projects launched,” but genuine adoption. Who uses AI every day? Which teams went from pilots to building AI into the fabric of their work? Adoption numbers early on set your baselines. They show you whether AI made it out of the hackathon and into production.
Next comes productivity, where the impact starts to show., This is where you can measure tangible gains. Look for time saved: faster project turnarounds, shorter response times, quicker proposal cycles. Every hour saved here is proof that AI is actually making work better.
Finally, there’s integration. This is when AI stops being a set of experiments and starts becoming infrastructure. The systems are interconnected, he data flows freely, governance evolves. This is when local wins compound into enterprise-wide transformation.
Companies that reach this stage see the biggest payoff. They’re more than twice as likely to realize measurable value.
But here’s the catch: you can’t skip steps. Teams that jumped straight to measuring ROI without first building adoption or and monitoring individual productivity indicators ended up frustrated. They were trying to prove impact without understanding where it’s coming from.
Building Better Guardrails
One of the most significant blockers to AI adoption is governance. Most leaders worry about AI misuse, and for good reason, but locking things down too tightly grinds innovation to a halt. Successful organizations are striking a balance between letting people tinker and maintaining control where necessary..
This means a few things in practice: developing approval gates for sensitive tasks. Establishing clear data policies. Maintaining visibility into who’s deploying which models, where.. The goal is to make secure experimentation the norm, not require a permission slip for each automation.
One financial services CIO described his strategy as“governed freedom.” He set up a structured environment where every individual in the organization could innovate using AI inside defined parameters. When something worked, IT helped scale it safely. They’ve deployed hundreds of new automations while staying compliant and accelerating the business.
What’s Next
The next generation of AI measurement won’t come from forcing AI into yesterday’s frameworks. It required recognizing AI’s value curve. How progress flows from individual adoption to team productivity to enterprise-wide impact.
This means paying attention to early signals, even when they look “soft” on paper. Daily active users. Workflow launches. Cross-system integrations. These forecast ROI more accurately than most companies understand, showing not just what’s been spent, but what’s starting to scale.
And here’s the twist: the companies making real progress aren’t the ones with the most sophisticated models or the biggest budgets. They figured out how to encourage experimentation, measure what matters, and create value without increasing risk.
Remember that CIO from the beginning? He stopped chasing a single ROI number. Instead, he measured adoption rates, signs of increased productivity, and levels of integration. Six months later, when the board asked again, he had an answer. Not just a single number, but a complete story of how AI was transforming the business..
