Book a call

22 Sept 2026•6 min read

Companies get $3.70 back for every dollar spent on AI, the best ones $10.30.

In short: yes, for most companies that measure it. In a survey of over 4,000 business leaders, companies got back $3.70 for every dollar spent on generative AI, and the leaders got $10.30. Another survey of 3,466 executives found that 74% earned their money back within the first year. Both studies were paid for by AI vendors, so the most reliable number is the one you measure yourself.

Return on AI is the question every owner asks first, and the answers online range from “it changes everything” to “most projects fail”. Both are true for someone. The useful work is reading what each study measured, who answered and who paid for it.

Key numbers

Result What it is Who measured it
$3.70 back per $1 spent, on average Survey of 4,000+ business leaders IDC, paid for by Microsoft, 2024
$10.30 back per $1 at the best companies Same survey, top group IDC, paid for by Microsoft, 2024
74% earned their money back within a year Survey of 3,466 executives at firms with $10M+ revenue Google Cloud, 2025
132% to 353% return over three years Projection for firms with up to 300 staff Forrester, paid for by Microsoft, 2024
40% less time on writing tasks, 18% better quality Randomised experiment with 453 office workers MIT, published in Science, 2023
25% faster work, 12% more tasks done Randomised experiment with 758 consultants Harvard Business School, 2023
$3.5 billion saved since 2023 IBM’s own back office IBM annual report, 2025
1.7x faster revenue growth for AI leaders Survey across companies Boston Consulting Group, 2025

How much do companies get back from AI?

The most quoted figure comes from IDC, a research firm, in a study paid for by Microsoft. Over 4,000 business leaders reported $3.70 back for every dollar invested in generative AI. The top group reported $10.30.

Google Cloud asked 3,466 senior leaders at companies with more than $10 million in revenue. 74% said they got their money back within the first year. Among those seeing productivity gains, 39% said staff output had at least doubled.

For small firms, Forrester projected a return of 132% to 353% over three years for companies with up to 300 staff using Microsoft 365 Copilot. Forrester calls it a projection, and Microsoft paid for it.

Which AI return numbers can you trust?

Sort them into three groups:

  1. Experiments. Researchers split people into two groups, one with AI and one without, and timed the same tasks. At MIT, 453 office professionals finished writing tasks 40% faster with ChatGPT, and independent graders rated the work 18% better. At Harvard Business School, 758 BCG consultants using AI finished 12.2% more tasks, 25.1% faster. These are the most reliable numbers, and each covers one type of task.
  2. Company results. IBM reported in its annual report, filed with the US regulator, about $3.5 billion in savings from running its own back office with AI. Klarna cut its cost per customer service case by 40%. These are measured, on real work, by the company itself.
  3. Surveys and projections. IDC, Google Cloud and Forrester asked leaders what they saw or modelled what they expect. They show the direction. Treat the exact multiples with care, especially when the study’s sponsor sells AI.

Why do some companies get 10 times more than others?

Boston Consulting Group found that only 5% of companies get large value from AI, while 60% report almost no revenue or cost gains. The leaders grow revenue 1.7 times faster than the laggards.

McKinsey points to the difference: nearly three quarters of the companies earning most from AI redesigned their workflows around it, against one quarter of everyone else. Giving staff a chatbot changes little. Rebuilding the procedure, so the AI does the routine steps and people make the decisions, is where the return comes from.

How to calculate the return for your own company

You need four numbers for one procedure:

  1. Hours per month your team spends on it today.
  2. Cost of an hour of that work, including salary taxes.
  3. Hours per month after automation, measured for at least a month.
  4. What the AI costs per month: the software, the model usage and the setup spread over a year.

Return = (hours saved × cost of an hour) − AI cost. If an order desk saves 60 hours a month at €25 an hour, that is €1,500 a month before AI costs. Start with the procedure your team repeats most often, because the hours add up fastest there.

Sources

See the future of your company operations

Book a call