Raising productivity with AI
Labour productivity growth in the Netherlands fell from 2.4 to 0.4 percent a year over fifty years. Working more hours does not fix that. Working smarter does, and AI can help if you know where it fits.
In an opinion piece on Joop, economist Antonie Kerstholt argues that the Netherlands should not work harder but become smarter. The figures below show why that is more than an opinion.
The problem in three figures
Productivity keeps growing more slowly
Between 1974 and 1983 labour productivity grew by 2.4 percent a year on average. Between 2014 and 2023 it was 0.4 percent. Each hour worked produces barely more than ten years ago.
Shortages sit in specific occupations
In 32 of 113 occupational groups the labour market is already tight and will get tighter up to 2030, from engineers to nurses. The CPB expects the market to adjust elsewhere. So it is not tight everywhere, but where it is, hiring more people is hard.
Small businesses use AI the least
In 2025, 66.2 percent of large companies used AI, against 29.8 percent of SMEs and 13.8 percent of micro businesses. Exactly where every hour counts, it is used the least.
What controlled research shows
A lot is claimed about AI and productivity. Below are only studies that actually measured the effect, with the limits the researchers name themselves.
Customer service: 14 percent more resolved per hour
Across more than 5,000 customer service agents, AI assistance resolved 14 percent more conversations per hour on average. For the least experienced agents it was 34 percent.
The limit: the most experienced agents gained nothing, and quality dropped slightly.
Consultants: faster and better, within the limit
In an experiment with 758 consultants, participants using AI completed 12.2 percent more tasks, 25.1 percent faster and with over 40 percent higher quality.
The limit: on a task outside what AI can do, they reached the right answer 19 percentage points less often. There, AI made things worse.
Writing: 40 percent less time
Among 453 highly educated professionals, writing tasks took 40 percent less time with ChatGPT, at 18 percent higher quality.
The limit: these were short, well-defined writing tasks in an online experiment.
Programming: faster, but not automatically
Developers using GitHub Copilot built a well-defined assignment 55.8 percent faster.
The limit: experienced developers working in their own large projects were 19 percent slower with AI, while believing they were faster.
What that means for your business
- The gain is in well-defined, recurring work, and it is largest for people who do not yet do a task every day.
- Outside its range, AI makes the work worse. If you do not know where that limit is, you lose more than you gain.
- Time saved only pays off if you spend it on something else. In Denmark, two years after ChatGPT arrived, AI use had no measurable effect on wages or hours.
- Feeling faster is not proof. Measure before and after, or you will not know.
How we approach it
We do not start with a platform but with a task. Which work takes the most hours now and keeps coming back? We measure that first. Then we put AI next to it, with a person checking, and measure again. If it works, we expand. If it does not, you learned that for little money.
Your data and your code stay yours, and where needed the AI runs in your own environment.
Sources
Where every figure on this page comes from.
- CBS: arbeidsproductiviteit over vijftig jaar · 2,4 procent per jaar in 1974-1983, 0,4 procent in 2014-2023 (2024)
- ROA en UWV: krapte per beroep tot 2030 · 32 van de 113 beroepsgroepen krap en krapper
- CPB: de arbeidsmarkt werkt ook bij flinke krapte · Column van de directeur, november 2025
- CBS: gebruik van AI-technologie naar bedrijfsgrootte · 66,2, 29,8 en 13,8 procent in 2025 (2026)
- Brynjolfsson, Li en Raymond: Generative AI at Work · NBER working paper 31161, 2023
- Dell'Acqua e.a.: Navigating the Jagged Technological Frontier · Harvard Business School en BCG, 2023
- Noy en Zhang: productiviteitseffecten van generatieve AI · Science 381, 2023
- Peng e.a.: GitHub Copilot en de snelheid van programmeren · arXiv 2302.06590, 2023
- METR: AI-tools bij ervaren open-source-ontwikkelaars · arXiv 2507.09089, 2025
- Humlum en Vestergaard: AI-chatbots en de arbeidsmarkt · NBER working paper 33777, 2025
- Kerstholt: Nederland moet niet harder werken maar slimmer worden · Opiniestuk, Joop
Our approach
From advice to management, with one partner
The same rhythm for every project.
Advice
We start with a conversation, not with code. First clarity on what you need and what you do not.
Build
We build and integrate to measure, with technology that fits you. You own your data and your code.
Manage
We stay involved: monitoring, adjusting and growing with your business.
Which work takes you the most hours right now?
In one conversation we look at whether there is a gain to be had, and how to measure it.
Call 085 083 5775A 30-minute intake