The AI Earnings Signals Expert Traders Need To Watch

I received a question from a reader this week that I thought was interesting: With so many companies talking about artificial intelligence, how do I know whether AI is actually improving a business?

Before ChatGPT was released in late 2022, about 1 in 5 companies in the Nasdaq 100 discussed AI on earnings calls. Today, nearly 4 in 5 do, according to research from the Nasdaq Economic Institute.

But hearing management talk about AI is not especially informative. As traders, we need to understand what sits behind their claims, particularly when AI is being used to explain changes in productivity, costs, margins, or future growth.

Companies are experimenting with measures such as token usage, API calls, tasks completed without human intervention, and hours saved. These numbers give management new ways to describe AI activity, but they are not equally useful when we are trying to assess what is happening inside the business.

Hours saved is a good example.

A 2026 study from the Bank of Korea found that generative AI reduced average work time among workers in South Korea by 3.8%, or approximately 1.5 hours per week. If all of that saved time had been redirected to productive activities, the researchers estimated a potential productivity gain of approximately 1%.

However, the researchers found that the relationship between time savings and actual output growth was essentially zero. AI was making individual tasks more efficient, but those gains generally had not translated into higher realized levels of productivity.

There were exceptions. The researchers found better results among self-employed workers, professionals, and intensive AI users. They concluded that organizational structure and performance incentives appear to influence whether AI generated efficiency gains translate into higher productivity.

This gives us a useful way to evaluate what we hear from companies. If management says AI saved employees thousands of hours, the next question is: what happened as a result? Did the company increase output without increasing headcount? Did costs decline? Did margins improve? Did the company remove a bottleneck that had been limiting growth?

The same applies to the number of tasks completed by AI. Suppose management reports one million automated tasks during the quarter. That number becomes more useful if we know what those tasks previously required and what changed after they were automated. Without that information, we have a precise statistic with very little context.

Token usage presents another issue. A 50% increase in token usage simply tells us usage increased. It does not tell us whether productivity, revenue, or profitability increased with it. Depending on the company, token consumption may become an important operating measure, but we still need to understand what that growth represents.

Traders have dealt with new metrics before. During the commercial internet era, some measures proved useful because they could be connected to economic outcomes. Others eventually disappeared. Nasdaq points to customer acquisition cost, customer lifetime value, and conversion rates as measures that endured, while “eyeballs” did not.

AI metrics will probably go through a similar sorting process. Until that happens, we have another layer of information to evaluate when companies report results.

I do not expect the useful measures to be identical across industries. A bank using AI to review fraud alerts should show progress differently from a manufacturer using AI to improve production. The metric needs to make sense for the job AI is performing.

For us, the practical question during an earnings report or company update is fairly simple: What changed because of AI, and can we see evidence of that change elsewhere in the results?

If management says productivity improved, we can look at output and labor costs. If AI is supposed to lower expenses, we can watch margins. If it is being presented as a source of growth, we can look for evidence in revenue, customers, or guidance.

That does not mean every AI benefit will immediately appear in quarterly results. It does mean we should be careful about treating a rapidly growing AI metric as evidence of improving performance until we understand what the number is measuring.

For traders, that is the useful part of this discussion. We are not trying to decide whether AI is an impressive technology. We are trying to determine whether the information companies are giving us about their usage changes what we know about the business, its expectations, and ultimately the trade.

Fortunately, I do not rely on AI to find trades. This isn’t because AI isn’t useful; it’s because I have the ITV indicator that finds them for us.