As precise as the railway is on time: German companies' demands on AI systems

76 percent of data managers from German companies have experienced business problems or crises caused by AI hallucinations or inaccurate AI responses in the past year. At the same time, 56 percent have abandoned a purchased AI agent system due to poor performance. 78 percent even believe that the company's management overestimates the accuracy of AI systems. This puts German companies well above the international average in all three areas - 59 percent for business problems, 45 percent for inadequate systems and 68 percent for the CEO's assessment of accuracy.

This is the result of a survey entitled Global AI Confessions conducted by the US market research company The Harris Poll. The survey was commissioned by Dataiku, a provider of a platform for AI and ML services. In 812 online interviews, 203 data managers from the USA, 103 from Germany, 103 from Japan, 102 from the UK, 101 from France, 100 from the United Arab Emirates, 50 from Singapore and 50 from South Korea answered the market researchers' questions. All participants work for large companies with an annual turnover of more than 1 billion US dollars or the regional equivalents. The survey participants work at the vice president, director, managing director, or C-suite level in data, analytics or AI-related functions.

Germans do not take it so seriously

Of the German respondents, only four percent required systems to be 95 to 100 percent accurate, compared to 15 percent internationally. 53 percent of Germans were satisfied with less than 80 percent accuracy in AI-supported decisions, compared to an international average of 38 percent. The market researchers see a link here to the highest rate of German projects failing due to low quality.

Just under 60 percent of respondents also stated that they are just waiting for a catastrophe caused by AI hallucinations in business-critical workflows. Just as many fear financial risks due to uncontrolled API and service costs caused by AI agents. At the same time, only 17 percent of German data managers demand complete end-to-end traceability of multi-agent workflows (globally: 20 percent). 34 percent of participants from Germany stated that their teams could track less than half of AI decisions in line with current regulatory requirements. 58 percent had even delayed or blocked AI agent deployments due to a lack of explainability. Only five percent of the total number of international respondents stated that they were able to record 100 percent of the decisions made by their AI systems.

Of the Germans, 63 percent fear that customers could be harmed by their AI agents, while as many as 69 percent believe that AI systems could harm their employees. The survey does not discuss possible security concerns in detail - it only recently emerged that even the industry leader OpenAI is unable to protect personal data from its users' emails from attackers. It concealed knowledge of a corresponding exploit for months.

Management overestimates the technology

In the press release for the survey, Dataiku speaks of a "gap between C-suite expectations and technical reality". This is particularly pronounced in Germany: Almost 80 percent of those responsible for data are convinced that their boardroom overestimates the accuracy of AI systems. Internationally, this impression was 68 percent. 82 percent of Germans stated that management underestimates the time and difficulty of bringing AI systems to production maturity. Of those surveyed, around 70 percent expect a CEO to be dismissed by the end of 2026 due to a failed AI strategy or an AI-induced crisis. On a global average, 56 percent of respondents expect this.

A similar survey of CEOs conducted in March 2025 revealed that almost 70 percent of 100 German CEOs also expect to be made redundant due to AI projects. The CEO survey also revealed that 95 percent of German CEOs have the impression that AI agents could provide equivalent or better advice on business decisions than a human board member. 76 percent of data managers share the view that AI-generated business recommendations carry more weight in their companies than those of human employees. Germany is also the survey leader here, with an average of 69 percent.

The surveys of CEOs and data managers can be found here and here. Access requires an email address.

(pst)