lltra software.
AI and Your Workforce: What the Research Actually Says
The debate nobody is winning
The conversation around AI and work has split into two camps: those predicting mass unemployment and those claiming AI will magically create more jobs than it destroys. Neither camp is paying much attention to the evidence.
At Ulltra, we help Swiss companies integrate AI into their operations. That means we need to be honest about what the data shows, not what makes for a good headline. Here is what the research actually tells us about productivity, job transformation, and retraining.
What the productivity data actually shows
The most rigorous studies paint a consistent picture: AI makes people meaningfully more productive, but the gains are uneven.
The landmark Harvard Business School and BCG study gave 758 consultants access to GPT-4 for realistic tasks. The results: consultants using AI completed 12.2% more tasks, finished 25.1% faster, and produced results rated 40% higher in quality. That is a real and significant effect.
Erik Brynjolfsson's Stanford study of 5,179 customer service agents found something equally important: novice workers improved by 34%, while top performers saw minimal gains. AI acts as a leveler, bringing less experienced team members closer to expert performance.
A 2026 Federal Reserve Bank of Atlanta working paper, surveying corporate executives, found that large firms expect output per worker to be roughly 3% higher in 2026 due to AI adoption. Across controlled experiments, productivity gains typically range from 20% to 60% on specific tasks, with 15% to 30% in real world settings.
AI helps your weakest performers far more than your strongest ones. This means the biggest ROI often comes from teams that are struggling, not from optimizing those already performing well.
But there is a catch. BCG's own 2025 survey found that productivity increases when people use three or fewer AI tools, and drops sharply when they use four or more. They called this "AI Brain Fry": 14% more mental effort, 12% more fatigue, 19% more information overload. More tools does not mean more output.
Which jobs change, which don't
The research consistently shows that AI augments knowledge work: writing, analysis, coding, customer support, and data processing. It has far less impact on work that is physical, deeply relational, or requires novel creative judgment.
An NBER study of U.S. labor market data through 2025 found no statistically significant decline in job openings or employment in AI exposed occupations. What changes is the composition of tasks within a role, not whether the role exists. By December 2025, 35.9% of U.S. workers reported using generative AI tools, concentrated among younger, college educated, and higher earning employees.
The honest answer: most jobs will change. Few will disappear entirely. The transition is gradual, not sudden.
What successful retraining looks like
This is where the evidence gets uncomfortable. Anthropic's August 2026 review of worker retraining programs found that traditional government retraining produces "small but positive impacts," with annual earnings increasing by roughly $1,100 in year two. That is not nothing, but it is far from transformative.
An analysis of 23 million records from the U.S. Workforce Innovation and Opportunity Act found that the program rarely shifted workers into less automation exposed occupations. Generic, classroom based retraining does not move the needle.
The one consistently strong outcome in the research? Employer led apprenticeship programs. Short, specific, on the job training that is tied to real work tasks outperforms generic courses by a wide margin.
This aligns with what we see in practice. The companies that succeed with AI adoption invest in structured, hands on training embedded in actual workflows, not in sending employees to generic "AI literacy" seminars.
What we recommend to our clients
Based on the evidence and our own project experience, here is what works:
- Start with willing teams. Forcing adoption creates resistance. Find the team that is curious, give them tools, and let their results speak for themselves.
- Measure before and after. Without baseline metrics, you cannot distinguish real productivity gains from enthusiasm bias. Track task completion time, output quality, and employee satisfaction.
- Limit tool sprawl. The BCG data is clear: three tools maximum. Pick the right ones, integrate them properly, and train deeply on those.
- Train on the job, not in the classroom. Build AI into existing workflows rather than running abstract workshops. Pair experienced users with beginners.
- Do not force adoption. Some roles and some people will not benefit from AI tools today. That is fine. Revisit in six months.
The Swiss context
Switzerland has advantages that most countries lack. The dual education system and apprenticeship tradition are exactly the kind of employer led, on the job training model that the research says works best. Swiss companies already know how to develop talent through structured, practical learning.
The Swiss labor market is also tight, with unemployment consistently below 3%. This means AI adoption here is less about replacing workers and more about doing more with the teams you already have. That is a fundamentally different challenge than what the U.S. or UK faces, and it requires different strategies.
Swiss companies should leverage their apprenticeship infrastructure for AI skill building. The model of learning by doing, supervised by experienced practitioners, maps directly onto what the research says about effective AI training.
The bottom line
The research supports a measured, evidence based approach to AI adoption. Real productivity gains exist, but they require thoughtful implementation, targeted training, and realistic expectations. The companies that will benefit most are not the ones rushing to deploy every new model. They are the ones investing in their people alongside the technology.
If you are considering AI adoption for your team and want to start with the evidence rather than the hype, get in touch. We help Swiss companies build AI strategies grounded in what actually works.