One of the most common arguments I hear for enterprise AI is productivity.

An employee uses AI and saves an hour.

A developer writes code twice as fast.

An analyst produces a report in 20 minutes instead of two hours.

All good things.

But I think we need to be careful about something:

Productivity is not necessarily ROI.

An hour saved isn't automatically an hour of economic value

Suppose I give 10,000 employees an AI assistant.

Each employee tells me it saves them four hours a week.

That's 40,000 hours of productivity every week.

Sounds fantastic.

But what exactly happened to those 40,000 hours?

Did the company generate more revenue?

Did we reduce costs?

Did we need fewer people?

Did we process more transactions?

Did we serve more customers?

Did we reduce risk?

Or did everyone simply have a little more time available during the week?

There is absolutely value in that. Employees may do better work, spend more time with customers, make better decisions or simply get through their workload faster.

But those benefits are much harder to put into an ROI calculation.

This isn't a new problem

We have seen versions of this throughout the history of enterprise technology.

A new system gets implemented and everyone agrees that it makes employees more productive.

Then someone asks:

Where is that productivity showing up in the financial results?

That's a much harder question.

AI makes it particularly interesting because the productivity improvements can be so dramatic.

If AI helps a developer complete something in two hours that previously took eight, we can measure the six hours saved.

But we haven't necessarily saved six hours of salary.

The developer still works for the company.

The economic value appears only if we do something useful with the capacity we just created.

Maybe the team builds more software with the same number of people. Maybe a product gets to market faster. Maybe we avoid hiring another 50 developers.

That's where productivity starts becoming economic value.

So perhaps we're measuring the wrong thing

A lot of enterprise AI measurement today focuses on usage.

How many employees have access?

How many use it every week?

How many prompts are they submitting?

How much time do they say they're saving?

Those are useful metrics when you're trying to understand adoption.

But they're not necessarily business outcomes.

As AI moves from experimentation into real enterprise deployment, I think the measurements have to move with it.

For some use cases, the answer will be revenue.

For others, cost reduction.

For others, increased capacity, faster time to market, fewer errors, reduced risk or better customer outcomes.

The metric will vary.

But eventually there has to be an outcome that matters to the business.

And this brings me back to the economics

In my last article, I argued that AI itself may become extraordinarily cheap while enterprise AI remains expensive because of everything required to put it safely and reliably into production.

That makes this question even more important.

It isn't enough to show that AI makes people more productive.

We need to compare the total cost of deploying AI with the economic value the enterprise actually captures from it.

I believe AI will create enormous productivity gains.

The harder question is how much of that productivity enterprises will actually convert into measurable economic returns.

And with hundreds of billions being invested in the infrastructure behind AI, that distinction matters.

Saving time is valuable.

Turning that saved time into economic value is where the real ROI begins.