As enterprises move from AI copilots to embedded workflows and autonomous agents, the cost model is changing underneath them. A single user request can now trigger dozens, sometimes hundreds, of model calls, retrievals, retries, and orchestration steps. A tool that looked affordable in a pilot can behave very differently in production. In a new feature on CIO.com, ModelOp CTO Jim Olsen joins other enterprise technology leaders to explain how CIOs can get ahead of runaway AI spend, and why the harder problem isn't the bill at all.
Use the right model for the job
One of the fastest ways enterprises overspend, Jim explains, is defaulting every task to the most powerful model available. Many workloads don't need frontier-level reasoning, and paying for it anyway is pure waste.
"It's like hiring the most expensive engineer to change a few colors in a website's CSS, or visual styling. You wouldn't do that. You use the appropriate tools for the task."
Jim Olsen, CTO, ModelOp
The discipline of routing work to the least expensive model that can still accomplish the business goal is one of the clearest levers CIOs have to bring costs down without sacrificing outcomes.
The deeper problem: value shock, not bill shock
For Jim, the bigger issue facing enterprises isn't the size of the AI bill. It's whether that spend is actually returning value. Spending $200,000 in a quarter is easy to justify if it produces $2 million in business value. The danger is pouring money into use cases that don't deliver a meaningful return.
"Are you actually getting that return on investment, or are you just blowing tokens for something that's not delivering the value to your business?"
Jim Olsen, CTO, ModelOp
As Jim puts it in the article, the question isn't whether someone used a million tokens. It's what they used them for.
Tie models and spend to a business use case
Tracking token usage by user or department shows who consumed AI, but not whether that consumption mattered. As Jim points out, the same underlying models and agents might power HR document search, customer support, code review, and problem resolution. Each of those use cases can rely on the same models, yet deliver very different business value.
That is why, for most enterprise AI systems, costs should be tied back to the business use case they serve, not just the user or team that ran them. Every model and every dollar of spend needs to be associated with a specific business use case. Only then can an organization see which spending is producing value and which is not.
Making that connection requires an AI inventory: a record of which business workflows use which models, agents, providers, and systems. Without that system of record, organizations can't associate spend with a use case, can't tie consumption back to value, and can't tell which AI spend deserves to scale.
Why this matters
Jim's perspective in CIO.com reflects the core of what ModelOp delivers for enterprise AI leaders. Managing AI cost, value, and risk at scale isn't a dashboard problem or an accounting exercise after the money is already gone. It requires a system of record that connects every model, agent, and workflow to the business use case it serves, and to the value it produces. That's how CIOs move from reacting to AI bills to actively governing AI value.
