In my essay The AI Dividend Fallacy, I argued that AI productivity gains do not automatically translate into lower costs. That naturally leads to another question I hear from executives:

How should organizations measure the ROI of Enterprise GenAI?

My answer surprises many people.

For general-purpose AI assistants such as ChatGPT, Claude, or Gemini, organizations often spend far too much effort trying to measure something that is not worth measuring precisely.

The economics are already favorable

In one organization I know well, Enterprise GenAI paid for itself if it improved employee productivity by less than 0.5%.

Think about that for a moment.

How would you reliably measure a 0.5% productivity improvement across thousands of employees performing hundreds of different knowledge tasks?

Building instrumentation capable of detecting a change that small can easily cost more than the uncertainty you are trying to eliminate.

Knowledge work is difficult to measure

There is a second problem.

Measuring productivity in knowledge work is far harder than many organizations assume.

Consider a simple thought experiment.

What productivity gain does your organization receive from Microsoft Office?

Do you calculate its ROI every quarter?

Most organizations do not.

They recognize Microsoft Office as enabling infrastructure rather than a narrowly defined business application.

Enterprise GenAI is increasingly becoming the same kind of capability.

Good governance still matters

None of this suggests that Enterprise GenAI should become a free-for-all.

Organizations still need:

  • Spending controls
  • Token caps
  • Employee training
  • Appropriate governance
  • Risk management

Those disciplines remain essential regardless of whether precise ROI is measured.

Measure where it matters

Rigorous ROI measurement becomes far more valuable when AI is embedded within repeatable, high-volume business processes.

Those systems are:

  • Easier to instrument
  • Easier to benchmark
  • Easier to compare against historical performance

More importantly, they often generate substantially larger measurable business improvements than broad productivity tools.

Organizations should focus their measurement efforts where they can produce meaningful insight rather than statistical noise.

The goal is not to avoid measurement. It is to measure where measurement creates value.

That distinction becomes especially important when AI moves beyond general-purpose assistants and becomes part of the organization’s core business processes.