Data and method
How we make the forecast
Same rules for everyone. Same profile, same result.
Sources
| Source | Markets | What we use |
|---|---|---|
| Arbeitnow | Germany and Europe | Counts, and title with link back |
| Remotive | Remote jobs | Counts, and title with link back after 24 hours |
| Adzuna | DE, AT, CH, NL, PL, GB, US | Counts only, after a commercial agreement |
Counting
- Every night we fetch new job posts from each source.
- We remove duplicates with a hash of title, company and country. The same job on two boards counts once.
- We tag each post with skills and a role, using a fixed list of more than 100 skills and their common names.
- We add the post to weekly counts per country, skill and role. Then we drop the text.
Forecast
- Demand = share of all job posts in a market that mention the skill, smoothed over 4 weeks.
- We fit a trend on the last 26 weeks, damp it, and project 6 months ahead. The range is an 80% band.
- Weather: Sunny above +10%, Partly sunny +3 to +10%, Cloudy -3 to +3%, Showers -15 to -3%, Storm below -15%.
- Confidence is low when a skill has few posts or few weeks of data. We show it next to every number.
Replaceability score
0 to 100, lower is better. It mixes three parts:
- 45% automation exposure of your skills, based on research on AI and task automation (ILO 2025, OECD, Eloundou et al. 2023).
- 30% skill durability, from the half-life of each skill.
- 25% market demand for your skill mix in your country.
Age, gender and photo never change the score.
Limits
- Job boards do not cover every job. Trades and public sector jobs are under-counted online.
- A skill in a post is not a hire. Counts show what employers ask for, not what they pay.
- Forecasts extend recent trends. Shocks like a recession are not in the model.
See the job market page for the current numbers.