# The AI figures that struck us most

Author: [Federico Tropea](https://it.linkedin.com/in/federico-tropea-415240266)
URL: https://www.goquadra.ai/risorse/articoli/i-dati-sull-ai-che-ci-hanno-colpito
Published: 2026-09-30
Updated: 2026-09-30
Language: en
Reading time: 6 min

Four figures from three 2025 studies, and what they tell us when read all the way through.

## At a glance

- Anthropic estimates 80% less time spent on tasks, excluding subsequent review.
- In BCG's study, the most advanced companies grow 1.7 times faster and have operating margins that are 60% higher.
- MIT NANDA links success to specific problems, integration into processes and external partners.

**Four figures, three studies**

| Figure | What it measures |
| --- | --- |
| −80% | Time spent on individual tasks<br>Anthropic · November 2025 |
| 1.7× | Revenue growth in companies furthest ahead with AI<br>Boston Consulting Group · September 2025 |
| +60% | Operating margin<br>Boston Consulting Group · September 2025 |
| 2× | Likelihood of reaching production with an external partner<br>MIT NANDA · July 2025 |

We start with the smallest scale, the individual task, and move to the largest: how to get there.

## −80%: the individual task

Anthropic, the company that develops Claude, has a vantage point almost nobody else has: millions of people using AI for work every day. Instead of a laboratory experiment, it analysed 100,000 real conversations.

The result: tasks that would take an average of around 90 minutes without AI are completed 80% faster. Within that average, some cases are even more striking. Gathering and synthesising information from multiple reports: around 95% less time. Preparing invoices, memos and documents: 87%. A curriculum that would take four and a half hours, written in eleven minutes.

These are estimates made by the model, not stopwatch measurements, and they do not count the time spent reviewing the work afterwards. Anthropic states this and points out that controlled experiments in previous years found smaller savings. But even the more cautious estimates suggest hours given back every week.

There is also a nuance that makes the figure even more interesting. **Writing a report 80% faster does not mean working 80% less: it means having time for the work that matters.** Anthropic itself observes that the biggest leaps in productivity have never come from completing old tasks faster, but from reorganising work around new possibilities. And that is exactly what the second study shows.

## 1.7× and +60%: the whole company

BCG interviewed 1,250 executives at large companies around the world and assessed the companies across 41 AI capabilities. It then compared the most advanced 5% with the 60% that are not yet deriving value from AI.

The former grow 1.7 times faster. This does not mean generating 70% more revenue, but growing faster: where one grows 5% a year, the other grows 8.5%. It seems small until you look at it over time. In ten years, the first grows 63%. The second more than doubles.

Operating margin, meanwhile, is 1.6 times that of companies that are standing still, meaning 60% higher. Proportionally, not in percentage points: if a company that is standing still has a 10% margin, an advanced one is at around 16%. Six points that, on revenue of one hundred million, are worth six million a year.

You could object that the best companies tend to do many things better, and that AI is not the only reason for the gap. That is true. But BCG also shows that the gap is widening: those ahead invest more, gain more and reinvest, in a self-reinforcing cycle. **Those who use AI well do not just have an advantage. They have an advantage that grows.**

And a figure from the same study tells us where that advantage comes from: 70% of AI's value is concentrated in the core business. Not in support functions, but at the heart of the company: how it designs, how it produces, how it sells, how it delivers.

It is the same conclusion Anthropic reached, by a different route. AI delivers more when it stops being a tool for completing peripheral tasks faster and becomes part of how the company creates value. And this is also the most exciting part: it means the greatest room for improvement is not in the inbox, but in the very work that makes a company what it is.

## 2×: how to get there

MIT NANDA analysed more than 300 business AI projects, using interviews and questionnaires. It is the report that became famous for a stark figure: 95% of projects do not generate measurable returns. A much-discussed figure, but the heart of the report lies elsewhere: in the 5% that succeed, and why.

The companies that succeed share similarities. They start with a specific problem rather than a technology. They bring AI into everyday processes, not alongside them. They choose systems that learn and adapt to how people work. And they often do not do it alone: projects developed with an external partner reach production around two times out of three, while those built entirely in-house do so one time out of three.

This figure is close to home for us, because an external partner is what we are. We chose it because it is one of the few numbers that speak to how an AI project is built, not just how much it returns. The most likely explanation is also the simplest: someone who has already seen a problem in other companies knows where it gets stuck.

## Taken together

Three studies, three different methods, one direction.

AI can already give back hours on almost every task. Companies that have turned those hours into a new way of working grow faster, earn more and widen the gap every year. And the path to get there is now visible: understand how the company really works, start with a real problem and implement AI exactly where it creates value.

For the first time, an advantage that once took decades can be built in a few years. The companies that lead their industries ten years from now will probably not be today's largest, but those that began rethinking their work first.

**The numbers say it is possible. The rest depends on when you start.**

## Sources and further reading

Anthropic, "Estimating AI productivity gains from Claude conversations", November 2025.

Boston Consulting Group, "The Widening AI Value Gap: Build for the Future 2025", September 2025.

MIT NANDA, "The GenAI Divide: State of AI in Business 2025", July 2025. Coverage: Fortune, 18 August 2025.

- [Anthropic · Estimating AI productivity gains from Claude conversations](https://www.anthropic.com/research/estimating-productivity-gains) — Published 25 November 2025. Accessed: 2026-09-30.
- [Boston Consulting Group · The Widening AI Value Gap: Build for the Future 2025](https://www.bcg.com/publications/2025/are-you-generating-value-from-ai-the-widening-gap) — Published 30 September 2025. Accessed: 2026-09-30.
- [MIT NANDA · The GenAI Divide: State of AI in Business 2025 (PDF)](https://cloudelligent.com/wp-content/uploads/2026/02/v0.1_State_of_AI_in_Business_2025_Report.pdf) — Edition July 2025. Accessed: 2026-09-30.
- [Fortune · MIT report: 95% of generative AI pilots at companies are failing](https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/) — Published 18 August 2025. Accessed: 2026-09-30.