It's 2026, and the initial rush to adopt generative AI is meeting a sobering reality check within Russian enterprise engineering teams. A new study by Intellectual Analytics reveals that only 7-10% of pilot projects involving large language models and chatbots at major domestic firms have successfully reached full production. The remainder were quietly cancelled, remain stuck in testing, or are undergoing significant transformation.

Conducted between December 2025 and February 2026, the survey covered 50 large organizations across IT, finance, logistics, and industry. The data exposes a significant gap between experimentation and operational value. Approximately 30-40% of these initiatives failed to meet financial expectations because engineering teams struggled to integrate models into legacy business processes. One respondent described an attempt to fine-tune the Chinese Qwen model for legal operations. The assistant consistently delivered accuracy below 30%, making it unfit for purpose. Many projects, experts note, launched for publicity rather than utility.

Despite Russian companies spending 90.3 billion rubles on AI implementation in 2024, with an average of 5.95 million per firm, half of the surveyed entities have pushed industrial deployment timelines from early 2026 to late 2026. The delay stems from unexpected complexity and a lack of immediate efficiency gains. Experts cite the inability of general models to handle specialized tasks, emerging security threats, and a shortage of qualified ML engineers as primary blockers. Corporate culture also lags behind technical capability. While bad actors use generated images to sell fake hybrid berries on marketplaces, serious engineering leaders are learning that base models do not solve niche problems without substantial customization. The path from proof-of-concept to production is proving far steeper than anticipated.

Source: Lenta.RU