Enterprise AI budgets are growing, but proving ROI remains difficult—especially once systems move beyond experimentation and into real production environments. In a new Computerworld feature, ModelOp CTO Jim Olsen explains why that breakdown is rarely caused by a single bad model or isolated misstep. More often, it reflects a structural problem: organizations build early AI initiatives in controlled settings, then struggle when real-world usage, cost variability, and operational complexity begin to scale.
Jim’s contribution stands out for its focus on what actually happens when AI leaves the lab. As he explains in the article, projects that look manageable during development can become much more expensive and unpredictable in production, where usage patterns shift, contexts expand, and true operating costs emerge. That challenge is even more pronounced with generative AI, where free-form interaction can drive unpredictable token usage and where models are reused across workflows and teams in ways that make both cost and value harder to attribute.
Jim notes that many organizations still lack a basic understanding of what AI systems they actually have in production. Without that visibility, enterprises cannot reliably measure performance, govern risk, or connect spend back to business outcomes. ModelOp is built to solve exactly that problem by giving enterprises a system of record for AI, along with the lifecycle oversight needed to track models, use cases, ownership, controls, and business value over time.
AI should be managed as enterprise infrastructure, not as a collection of disconnected experiments. Jim argues that lifecycle management—covering development, deployment, monitoring, and retirement—is not administrative overhead. It is what makes accountability possible as models evolve and usage grows.
For enterprise leaders, the takeaway is clear. AI ROI does not disappear simply because the technology stops working. It often erodes because the organization loses visibility into what is running, what it costs, who owns it, and whether it is still delivering value. Jim’s comments in Computerworld offer a practical reminder that better lifecycle governance is not separate from ROI discipline—it is one of the main conditions for achieving it.
