If you have built a reskilling business case in the last year, you almost certainly quoted a number about how much of work is about to change. It is worth knowing where that number came from, because the two kinds of evidence available point in noticeably different directions.
The survey number, and the thing nobody mentions about it
The most-quoted figure comes from the World Economic Forum: employers expect 39% of workers' core skills to change by 2030, and 63% name skills gaps as the single biggest barrier to transforming their business. That is from the Future of Jobs Report 2025, drawn from more than a thousand employers across 22 industries and 55 economies.
Here is the part that rarely makes it into the deck. That 39% is down from 44% in the previous edition. The number moved in the opposite direction to the story it usually gets used to tell. It is also employer opinion, not measurement. Nobody counted skills. Executives were asked to estimate, and estimates drift with the mood of the moment.
The same report gives a more useful breakdown. Out of every 100 workers, 29 are expected to be upskilled in their current role, 19 reskilled and redeployed, 41 to need no significant training at all, and 11 to need training they are unlikely to receive. That last group is the actual risk. It is a much smaller and more specific problem than "everything is changing."
What the hiring data says instead
Job postings are harder to argue with, because somebody had to write and pay for them. Stanford's 2026 AI Index reports that AI skills now appear in 2.6% of all US job postings. In the information sector it is 13.2%, nearly double the year before. Postings mentioning agentic systems grew by over 10,000% year on year.
Both halves of that matter. Growth is genuinely explosive, and it is explosive from close to zero. In 2026, roughly 97 out of every 100 American job postings still do not ask for AI skills at all. Anyone quoting the growth rate without the base rate is selling you something.
There is a real premium attached, though. Lightcast, analysing more than 1.3 billion postings, found roles listing AI skills advertise about 28% higher salaries, roughly $18,000 a year, and that 51% of those roles sit outside IT and computer science entirely. Two caveats: Lightcast sells labour market data, and advertised salary is not paid salary. AI-skilled roles also skew senior, so this is a correlation, not a return-on-training estimate.
The one piece of causal evidence I could find is more modest and more interesting. An experiment with 1,700 hiring professionals, reported by the WEF in February 2026, found candidates with AI skills were 8% to 15% more likely to be invited to interview. Smaller than the salary headline, but it is an actual experiment rather than a correlation.
The counterpoint worth taking seriously
The US government's own projections do not describe a collapse. The Bureau of Labor Statistics projects software developers growing 15.8% and data scientists 33.5% between 2024 and 2034. The declines are concentrated and specific: customer service representatives down 5.5%, legal secretaries down 5.8%.
Worth naming the limitation. A projection model built on historical relationships is not designed to catch a fast structural break, and the BLS piece says nothing about how it handles that uncertainty. But it is real government data, and it is not describing the world the survey narrative implies.
The one number that should actually worry you
While everyone argues about how much work is changing, formal learning is quietly shrinking. The Association for Talent Development's 2025 State of the Industry puts average formal learning at 13.7 hours per employee, down from 17.4 the year before. Direct spend was $1,054 per employee, and the cost per learning hour rose 34% to $165.
Fewer hours, more expensive hours, and a workforce being told everything is about to change. That is the tension to plan around. It is based on 2024 data from 539 self-reporting practitioners, so treat it as directional rather than precise.
What to do with all this
Stop building the case on the 39%. It is soft, it moved the wrong way, and a sceptical CFO will find that out. Build it on three things instead: the 11 in 100 who need training and will not get it, the specific occupations where the hiring data actually shows movement in your industry, and the fact that your own learning hours are falling while expectations rise.
That is a narrower argument. It is also one that survives contact with someone who checks.
On sourcing. Every figure above links to the study it came from. Two widely circulated numbers are deliberately absent: the claim that 70% of job skills will change by 2030, which traces to LinkedIn but has no published methodology, and a figure of 1.6 million unfilled AI positions globally, attributed to a report I could not locate anywhere. If a statistic cannot be traced, it does not belong in your business case.