Total Factor Productivity

When an economy produces more than it did the previous year, there can be two very different reasons. Either it simply deployed more: more hours of work, more machines, more capital. Or it used the same means more cleverly. Total factor productivity – TFP for short – measures precisely this second share. It is the portion of growth that remains once you strip out everything that can be explained by simply using more resources. Economists therefore sometimes call it the “measure of our ignorance”: whatever we cannot explain through labor and capital ends up in TFP.

TFP as a gauge of progress

Growth through more labor or more machines is finite. At some point the population is exhausted, at some point the next factory no longer pays off. Growth through higher TFP, by contrast, is in principle unlimited. That is why economic policy is so strongly interested in it.

A historical example makes this tangible: industrialization in the 19th century was not so powerful merely because suddenly more people worked in factories. The steam engine made it possible to produce many times more with the same number of people. That is TFP growth. Current research hopes for something similar from artificial intelligence: if AI systems help engineers, doctors, or programmers accomplish significantly more in the same amount of time, this shows up directly in TFP.

Conversely: stagnating TFP is a warning sign. Many economists explain the weak growth in Western countries since the 2000s by the fact that TFP has barely risen – despite immense investments in digital technology.

Calculation: What goes into TFP

The basic idea is simple: you take an economy’s total output and subtract the share that can be explained by the use of labor and capital. What remains is considered TFP. In practice, a formula called the Cobb-Douglas production function is used for this, which indicates how strongly labor and capital each contribute to output.

That sounds clean, but in practice it is tricky. How do you reliably measure “capital”? Do two old machines count as much as one new one that is twice as fast? And what about the workforce’s knowledge, which appears nowhere on the balance sheet? Such measurement problems mean that TFP figures from different countries are only comparable to a limited extent.

It is also important to note what TFP does not separate out: it contains technological progress, better organization, more education among the workforce, and even luck – such as particularly good weather for agriculture. It is an aggregate measure, not an X-ray of a single factor.

TFP in the AI debate

Since the breakthrough of large language models like GPT-4, TFP has regularly appeared in economic reports and tech news. The central question is: will AI measurably raise TFP – or will it remain an expensive tool without an economy-wide effect? So far the findings are mixed. In individual sectors such as software development or drug research, there are indications of genuine productivity gains. For the economy as a whole, the effects are not yet clearly measurable.

This recalls an earlier puzzle: in the 1980s, computers were already widespread, but TFP figures showed hardly any response. The economist Robert Solow put it pointedly: “You can see the computer age everywhere but in the productivity statistics.” Only in the 1990s did TFP clearly pick up. Many economists expect AI to have a similarly delayed effect – because companies must first restructure their processes before output increases.

So when you read that a tech company or a government wants to use AI to “boost productivity,” that almost always means: TFP. It is the decisive yardstick for whether a new technology truly changes something – or merely produces expensive headlines.