Governance and Observability Are The Accelerators That Turn AI Spend Into Real Returns

Most organizations still measure AI by the hours Copilot shaves off an individual's week. It's a tidy number, and it's also a trap. Counting saved minutes captures the first wave of the technology, the consumer-grade productivity gains, while missing the shift that actually moves the business: AI taking on strategic work across the enterprise. The organizations pulling real returns are treating governance and observability as the enabling layer that lets them trust what AI is doing, measure it honestly, and scale it with confidence.

Helping organizations shift to this approach is George Hynes, VP of Sales at G-Factor Consulting. He's an enterprise go-to-market and transformation leader whose recent work has centered on AI and automation in the life sciences space. Hynes has spent his career close to how large, highly regulated organizations actually adopt new technology, and in his view, too many firms view governance as a brake when they should be looking at it as an accelerator.

"If you use AI governance and observability as an enabler to scale AI, that's when you start to see the real returns," he says. Getting to those returns, though, means first understanding why the numbers most companies track today capture so little of what AI is actually worth.

The first wave was never the whole story

Hynes traces AI's arc from fixing grammar and drafting emails to running research and building macros, a steady climb that's reached genuinely strategic ground. On the consumer side, he sees people leaning on AI for everything from home projects to recording music to troubleshooting a broken TV. "People are using AI almost as a personal guru," he says. This is what he describes as the first wave.

The corporate side is a different conversation entirely, and it is where the narrow ROI lens starts to fail. As AI moves into strategic roles, the question stops being how many minutes it saved one analyst and becomes how the organization manages it responsibly at scale. Hynes points to how fast that frontier is arriving in pharma, where companies are approaching the point of using AI to help determine which molecules combine into viable drug candidates before anything reaches a clinical trial. "When you get to that kind of scale, you really need to look at it not only from what AI is producing, but what's the ROI for managing it from this observability and governance standpoint," he notes.

To scale, you have to trust

That management question is exactly what Hynes has to answer when he is in a room of executives who want business outcomes, and he reaches for a single line. "To scale, you have to trust. If you don't trust, you can't scale," he says. Governance and observability act as a verification layer, but he's careful to separate the idea from surveillance. The goal is not a big brother watching employees, but a structured way to know what AI is doing, who approved it, and whether it's behaving as intended, security included.

He points to a conversation with a technology leader from a major pharmaceutical company who described a mature setup, one built on approval thresholds scaled to the stakes of each AI investment. "They have it pretty well figured out, depending on the impact and the size, who has to approve what before somebody can go make an investment or do a project," Hynes shares. The approvals are only half of it. "They're also using observability tools to make sure that what's supposed to happen is actually happening, and then on top of that, their smartest people are watching the observability tools." That structure, in his telling, is what lets a cautious enterprise move from "we have to do AI" anxiety to defensible, scalable deployment.

Why the productivity numbers deserve scrutiny

Hynes is direct about how much of the reported AI productivity gain is real and how much is optimism dressed as data. Some of it is straightforward math, he says, but more of it is politics. The math only works with a baseline. "If I don't know what the productivity is today, before AI, how can I legitimately show what the improvement is afterward?" Without a before-snapshot of a defined process, any gain is really a projection of what a team hopes the improvement will be.

The politics are thornier, and they cut against honest measurement. Teams that prove a large efficiency gain often find finance reallocating the freed-up budget elsewhere, which makes them reluctant to report success too precisely. "There's this fine line of, how do I show success, but then how do I do it in such a way where I don't lose my budget?" Hynes has watched organizations agree to speak in general terms about their AI progress while declining to get specific, for fear of losing their funding. It's a dynamic that predates AI, but the scale of current spending is bringing it to a head as the C-suite starts asking what its token bills are actually buying.

The measurement gap, and how to close it

For companies that let individual contributors adopt AI first and never built a framework, Hynes' advice is direct: start measuring now, even if it's late. He points to benchmarks from McKinsey, Deloitte, IBM, and Microsoft as a way to set expectations, but insists benchmarks are only a starting line. Microsoft's own data, he notes, suggests roughly 40 percent of deployed Copilot licenses are actually used regularly, while Deloitte research puts real-world usage closer to 70 percent when deployed, a spread that tells its own story about the gap between buying AI and using it.

That gap is why he pushes teams to define a process, time it before and after, and back into the savings from there. "Even if you have to put a stake in the ground and say, I don't know what it was when I first started, but I'm going to measure it right now and go forward with the measurement," he says. "If you don't measure it, your progress is only a guess." The selectivity is deliberate. Companies are increasingly unwilling to pay a premium per user for a general-purpose guru, and are turning the capability on only where a specific, measurable job justifies it.

Agentic AI raises the stakes on oversight

The reason to build this discipline now, in Hynes's view, is that the ground is about to shift again. Today's AI mostly fetches and assists while humans assemble the result, which is roughly the arrangement most enterprises want. Agentic AI changes the contract. "Agentic AI is going to say, give me an outcome that you want and I'm going to figure it out." Instead of answering specific prompts, it pursues a goal on its own and decides the steps needed to get there.

That autonomy is what he calls observability on steroids. Hynes references press accounts of advanced systems taking unintended actions to complete a task they were given, then reporting back only that the objective was met, as a sign of how much harder oversight becomes when AI acts independently. How those agents are identified, orchestrated, and governed becomes the central question. The organizations that treat governance and observability as the enabler of scale rather than a tax on it are the ones positioned to trust these systems enough to actually use them, and to prove the returns when the C-suite asks.