Skip to main content

Is Your AI Actually Creating Economic Value?

on

Is Your AI Actually Creating Economic Value?

AI adoption is easy to measure.

  • Users
  • Prompts
  • Tokens
  • Agents
  • Automations
  • Usage curves

Dashboards can make an AI program look extremely busy.

The harder question is: Did any of it make the business economically better?

That distinction may become increasingly important as companies move beyond experimentation and start asking AI to earn its keep.

Because adoption is evidence of activity. It is not evidence of value.

Busy is not the same as productive

A recent Harvard Business Review article, AI and the Looming Competition for Margin, makes the measurement problem unusually clear.

The authors argue that companies should eventually evaluate AI through actual changes in productivity rather than easy-to-count inputs such as token usage or lines of code. They point to one striking example: a more than tenfold increase in lines of code written was associated with only a 1.3x increase in product releases.

That is a useful warning.

AI can increase activity dramatically without producing an equivalent increase in economically meaningful output.

So the first question in our Create → Capture → Defend framework is deliberately simple:

Did AI make anything economically better?

What should count as value?

That answer will vary by business and workflow. AI might create value by:

  • lowering operating costs
  • increasing output
  • accelerating cycle times
  • increasing revenue
  • improving customer outcomes
  • reducing errors
  • lowering risk
  • increasing productive capacity

The point is not that every AI initiative must immediately increase profit. The point is that the organization should know what economic outcome it expects to improve.

Otherwise measurement drifts toward whatever is easiest to count. And AI provides plenty to count.

Tokens are measurable. Prompt volume is measurable. Adoption is measurable.

None of those answers the question: Are we better off?

Productivity is not automatically profit

Even productivity itself requires care.

Suppose an employee completes a task in 30 minutes instead of an hour. That is a real productivity gain. But what happens to the saved 30 minutes?

  • Does the employee produce more?
  • Does the company need fewer labor hours?
  • Does customer service improve?
  • Does revenue increase?
  • Does the workflow bottleneck simply move somewhere else?

The HBR authors make a broader version of the same argument: doing “the same with less” or “more with the same” does not necessarily translate directly into more revenue.

So: Productivity ≠ profit

A faster task is not automatically a more valuable enterprise. The gain has to travel through the workflow and eventually appear somewhere economically meaningful.

This is where AI ROI gets harder

An AI initiative should therefore begin with a measurable hypothesis. Something like:

If we deploy AI here, we expect processing cost per case to fall by 15%.

Or:

We expect sales capacity per representative to increase without reducing conversion quality.

Or:

We expect average resolution time to fall while customer satisfaction remains stable or improves.

Now there is something to test.

Without that, AI programs can become collections of impressive capabilities looking for financial explanations after the fact. That reverses the proper sequence.

The business outcome should define the measurement. The model should not define the business outcome.

Positive activity can hide weak economics

This becomes especially important because AI itself costs money: models, infrastructure, data engineering, software, integration, governance, people.

The HBR article opens by noting that executives are already under pressure to justify growing spending on data scientists, software, and tokens.

So an initiative can be technically successful and still fail economically. It may:

  • work well
  • save time
  • attract users
  • produce impressive outputs

and still cost more than the economic benefit it creates.

That gives us another distinction:

Useful AI ≠ valuable AI

Useful is a technical and operational judgment. Value is an economic judgment. Businesses need both.

The first management question

For every AI initiative, leadership should be able to ask:

  • What economic outcome was supposed to change?
  • What actually changed?
  • What did it cost?
  • What is the trend?
  • How confident are we that AI caused the improvement?
  • Is the value large enough to justify continued investment?

Those questions sound basic. They are not. They force the organization to move from AI enthusiasm toward operating discipline.

And that is the first stage of Create → Capture → Defend:

Create: Did AI actually make anything economically better?

But even if the answer is yes, the analysis is not finished. Because creating value does not mean your company got to keep it.

That is the next question: Who captured the AI productivity gain?