Reading time: 5 minutes
64% Have Tried AI – But Most Are Doing It in Secret: What New Research Means for Your Team
Two-thirds of your employees have probably already experimented with ChatGPT. Most of them did it without IT's blessing. A new study from Germany's ifo Institute reveals what many COOs suspect: AI has arrived in daily work – but it's chaotic, uncontrolled, and without measurable ROI.
Published on January 7, 2025
The Numbers: Plenty of Curiosity, Little Structure
The study "Low Barriers, High Stakes" analyzes data from nearly 10,000 employees in Germany. The findings are sobering:
| Metric | Value |
|---|---|
| Have used AI at work | 64% |
| Use AI regularly | 20% |
| Use AI on own initiative | 66% |
| Employer-introduced AI | 33% |
The gap between "tried it once" and "productively integrated" is massive. And here's the real issue: two-thirds of AI usage happens in the shadows – without official approval, without training, without governance.
The researchers call this "dual diffusion": AI spreads simultaneously from the top (through employers) and from the bottom (through employees). The problem: informal diffusion is faster but shallower – it reaches many, but without depth.
The Problem Behind the Numbers
When employees privately use ChatGPT to draft emails or prepare presentations, three risks emerge:
1. Data Privacy and Compliance
Sensitive company data ends up in public AI tools. GDPR violations are practically guaranteed. With the EU AI Act adding documentation requirements, "shadow AI" becomes a liability.
2. No Measurability
Without structured rollout, there's no data. How much time is your team actually saving? Which tasks are being automated? The answer: nobody knows.
3. Amplified Inequality
The study is clear: AI adoption is "skill-biased." Higher-educated workers use AI significantly more. Without targeted intervention, the gap widens between those who master AI and those left behind. Women are 6 percentage points less likely to have increased their AI usage.
What Structured Implementation Achieves
The ifo study also delivers good news: Where companies formally introduce AI, employees report:
- Higher usage frequency – nearly half of formal users apply AI "often" or "always," compared to only a quarter of informal users
- More training – 8 percentage points higher training participation
- Higher perceived productivity gains – measurably better outcomes in time, quality, and output
Importantly: these effects persist even when controlling for industry, occupation, and even individual companies. Formal introduction works – regardless of which firm you're in.
From Curiosity to Competence: What You Can Do Tomorrow
The 44-percentage-point gap between "tried AI" (64%) and "uses AI regularly" (20%) isn't a lost cause. These are employees who are ready – they just don't know how to continue.
Step 1: Take Inventory
Ask your team: Who uses which tools? For what tasks? You'll be surprised how much shadow AI already exists.
Step 2: Set Guidelines
What's allowed? What isn't? A clear policy removes uncertainty and creates legal clarity.
Step 3: Train in Small Doses
Full-day workshops don't work. The study confirms: complex applications don't reach the majority. What works are small, continuous learning sessions that fit into daily work.
Step 4: Measure Progress
Without metrics, there's no management. Define what "AI Fluency" means for your team – and track it.
The Real Signal from This Study
The ifo researchers put it precisely: "Widespread informal usage can coexist with limited productivity effects when complementary investments and organizational integration lag behind."
Translation: Many use AI, but without training, without integration, without support – and that's why the productivity boost fails to materialize.
64% curiosity isn't a weakness – it's untapped potential. The question is: Will you let it grow in the shadows? Or will you make it productive?
Teams looking to build AI Fluency systematically can explore AI Guru – a Slack-native micro-learning program: 5 minutes daily, 12 weeks of structure, measurable results.
Source:
Arntz, M., Baum, M., Brüll, E., Dorau, R., Hartwig, M., Matthes, B., Meyer, S.-C., Schlenker, O., Tisch, A. & Wischniewski, S. (2025). Low Barriers, High Stakes: Formal and Informal Diffusion of AI in the Workplace. ifo Working Paper No. 422. Munich: ifo Institute.