Your Business May Be Producing More Because of AI. That Doesn’t Mean It’s More Productive.

One of the easiest things to demonstrate with artificial intelligence is speed.

A report that took an hour to draft can be prepared in minutes. A marketer can produce dozens of campaign variations. A developer can use an AI assistant to generate code, explain unfamiliar systems, or investigate a problem.

These improvements are tangible. They make AI attractive to employees and business owners alike.

But speed creates a measurement problem.

When a business can produce more output at a lower cost, it is tempting to assume that productivity has improved. Sometimes it has. Other times, the additional output creates more work downstream, requires extensive correction, or has little effect on the result that matters.

A business can become remarkably efficient at doing things its customers do not need.

What the research actually tells us

A 2025 field experiment by researchers affiliated with Microsoft Research and Harvard Business School examined generative AI use among 7,137 knowledge workers across 66 firms.

Workers who used the AI tool frequently spent less time on email and reduced work outside regular hours. Yet the researchers did not detect corresponding changes in the quantity or composition of their tasks simply from providing individual access to AI.

That is an important distinction. Saving time at the individual level does not automatically change what an organization accomplishes. Source: National Bureau of Economic Research, Working Paper 33795.

The study does not establish that AI fails to improve productivity. It shows why the effects need to be examined at more than one level.

An employee may finish a task sooner. Whether the business uses that time to serve more customers, improve quality, or reduce costs is a separate question.

The hidden cost of more output

Imagine a company using AI to increase the number of proposals it sends to prospective customers.

The initial numbers look encouraging. The sales team produces more proposals, and each one takes less time to prepare.

But suppose the proposals are less carefully qualified. The team spends more time following up with unsuitable prospects. Managers have to review more documents. Conversion rates remain unchanged.

The company has increased output without necessarily improving commercial performance.

The same problem appears in software development, customer support, marketing, and administration. Faster production can be valuable, but its value depends on what happens to the work afterward.

There is also a less visible cost: verification. If an AI system produces an answer quickly but requires a person to check every detail, the relevant measure is the time required to complete the task reliably, not the time required to generate the first draft.

Four questions worth asking

Before declaring an AI implementation successful, business leaders should establish what they expect it to change.

• Time: Has the total time required to complete the workflow actually fallen?
• Quality: Are the results at least as accurate, useful, and consistent as before?
• Economics: Does the improvement justify software, integration, training, oversight, and correction costs?
• Business outcomes: Has the business improved something that matters, such as customer retention, delivery time, profitability, or its ability to serve more customers?

The answers will differ by application. Some tools primarily improve employee experience. Others may reduce costs or enable entirely new services. Both can be worthwhile, but they should not be confused.

Productivity is a management decision

AI can create spare capacity. It does not decide how that capacity should be used.

A business that saves ten hours a week could use that time to improve its service, develop new products, deepen customer relationships, or simply reduce employee workload. Each outcome has value, but the business needs to be intentional about which one it wants.

This is where the conversation about AI should become more serious.

The question is not whether the technology can produce more. Increasingly, it can.

The question is whether the organization has the judgment and operating discipline to turn that additional capacity into something worthwhile.

Otherwise, we risk building businesses that generate more work, measure more activity, and still struggle to make meaningful progress.

That is not a technology failure. It is a failure to define what productivity is supposed to achieve.