In the early 1900s, factories across America were electrifying.
Business owners knew electricity was the future. It was cleaner than steam power, easier to distribute, and promised dramatic efficiency gains. Investors funded modernization projects, factories installed electric motors, and executives confidently predicted a new era of productivity.
Then something strange happened.
For years, productivity barely moved.
Economists would later call this the productivity paradox. Businesses had adopted a revolutionary technology, yet the numbers refused to cooperate. The problem wasn't electricity itself. It was how companies thought about it.
Most factories simply replaced a steam engine with an electric motor and continued operating the same way they always had. Electricity was treated as a substitute rather than a catalyst for change. The real gains came later, when manufacturers redesigned factory layouts, workflows, and production systems around what electricity made possible.
Only then did productivity explode.
I've been thinking about that story a lot lately because finance teams are starting to have a similar conversation about AI.
The first phase of AI adoption was easy. Engineers wrote code faster, support teams experimented with agents, and marketing teams generated content at unprecedented speed. AI lived inside innovation budgets where experimentation mattered more than measurement.
Now those invoices are becoming operating expenses.
And operating expenses invite tougher questions.
Most finance leaders probably want answers to a few simple questions: Which AI initiatives consumed the budget? Which generated the most value? If spending needed to be reduced, what would be cut first?
The challenge is that many organizations can't answer those questions with confidence.
Part of the problem is that AI reporting focuses on activity. We count users, prompts, tokens, and seats because those metrics are easy to collect. They tell us who adopted the technology.
They don't necessarily tell us whether the business improved.
A company can have thousands of employees using AI every day and still struggle to explain its impact on revenue, profitability, customer retention, or operational efficiency. Adoption and ROI often move together, but they are not the same thing.
The electricity story highlights the distinction. Measuring success by the number of electric motors installed would have told factory owners very little about whether productivity improved. What mattered was output.
The same principle applies to AI.
One of the more interesting examples came from Klarna. When the company announced that its AI assistant was performing work equivalent to hundreds of customer service agents, the headline attracted debate. What interested me was the measurement approach. The company attempted to connect AI to service volume, response times, inquiry handling, and expected financial outcomes.
In other words, it connected spending to output.
That's a very different conversation than talking about prompt volume or active users.
Engineering teams might measure cost per shipped feature. Support organizations could look at cost per resolved ticket. Sales teams might evaluate opportunities influenced by AI-generated work. The exact metric matters less than creating a relationship between the invoice and the outcome.
This becomes even more important as AI moves beyond assistance and closer to execution.
Recently, Robinhood announced plans that would allow AI agents to participate more directly in trading workflows. Whether that specific implementation succeeds is less important than what it signals. AI is gradually becoming operational infrastructure rather than productivity software.
At that point, usage becomes a less useful measure of value.
Nobody evaluates electricity by counting how often employees flip light switches. They evaluate what the factory produces. I suspect AI is heading toward the same destination.
The companies that benefit most from AI won't necessarily be the ones spending the most money. They'll be the ones that understand what their spending produces.
Because once a technology becomes essential, the question is no longer how often people use it.
The question becomes: what would happen if it disappeared tomorrow?
That's usually when you discover its real value.