---
title: "AI is in almost every company; measured productivity is not there yet"
slug: estudio-productividad-ia-medida-ejecutivos
lang: en
situation: medir
format: estudio
minutes: 5
updated: 2026-10-19
publisher: "NBER (National Bureau of Economic Research)"
published: "2026-10-01"
sources:
  - label: "NBER, Working Paper 34836 \"Firm Data on AI\" (October 2026 revision)"
    url: "https://www.nber.org/papers/w34836"
    date: "2026-10-01"
---

# AI is in almost every company; measured productivity is not there yet

> For anyone reporting AI's productivity effect to leadership who needs to set expectations before the next budget.

## In 30 seconds

- A survey of nearly 9,000 CEOs, CFOs and senior executives in seven countries, including Mexico, finds that 78% of firms use AI, yet the reported contribution to productivity over the last three years is 0.2% a year.
- If your business case promises large productivity jumps within months, it runs against what executives themselves report.
- Measure against a baseline and keep usage and impact apart in your reporting.

## The exhibit

```exhibit
title: AI's contribution to productivity is small and expected to grow
unit: %
Last 3 years (reported) | 0.2
Next 3 years (expected) | 0.6
note: NBER Working Paper 34836, "Firm Data on AI", Oct. 2026 revision. Nearly 9,000 executives in the US, UK, Germany, Australia, Canada, Mexico and South Africa; Summer 2026 survey. Annual labor productivity growth attributed to AI.
```

## Fact

- 78% of surveyed firms use AI. 80% of executives use it regularly, averaging 2.1 hours a week, up from 1.5 hours in Winter 2026.
- US firms expect to spend $2,000 per employee on AI in 2026. Spending is highly concentrated: 2% of firms account for over half of it.
- Net effect of AI on employment over the last three years: essentially none. More than 40% of firms expect positive or negative effects that roughly cancel out.
- Productivity: executives attribute 0.2% annual growth to AI over the past three years and expect 0.6% a year over the next three.

## My read

High adoption with low effect is not a contradiction. It is the gap between using a tool and changing how the work gets done. An executive using AI two hours a week does not redesign a process. For a manager, the practical lesson is about expectations: if the market expects 0.6% a year, promising 30% productivity in one quarter (an illustrative figure) leaves you exposed. It is easier to defend one concrete process, with its baseline, where the change can actually be measured.

## What to do on Monday

1. Before you promise productivity, define what you will compare against and with which data point from today. [Sheet: baseline, compared to what?](/resources/linea-base-comparado-con-que/)
2. In your next report, split two columns: usage (how many people use it) and impact (which business metric moved). [Sheet: hours vs. growth scorecard](/resources/scorecard-horas-vs-crecimiento/)
3. Check where your AI spend goes. If a few use cases take almost all of it, decide which to keep, adjust or switch off. [Sheet: keep, adjust or switch off](/resources/seguir-ajustar-apagar/)

## What we don't know

- The figures are executives' perceptions, not productivity measured in financial statements or national accounts.
- It is a working paper without peer review; the October revision changed the sample and figures from the February version.
- The abstract does not break results out by country, so we don't know how far Mexico sits from the average.

## If this becomes a big decision

If the decision is hard to reverse and carries your name, the free tools at vicsol.org/decision-clarity/ help you structure it. They don't replace a formal review.

---

Vic Sol · Decision Desk · vicsol.org/resources/ · Free to use with attribution.
Independent summary of a public source; figures belong to their author. Not legal, financial or audit advice.