Businesses Disappointed with Generative AI Implementation Early On

Despite significant investments and high interest in generative AI, most corporate implementation projects remain in the pilot phase. Experts attribute this to integration complexities, a lack of expertise, and the immaturity of current solutions. Widespread adoption is only expected in the coming years, as the market evolves and practical experience accumulates.

The share of pilot initiatives related to generative AI implementation remains extremely limited. According to a survey by 'Intellectual Analytics,' only about 7-10% of projects launched by major tech companies in 2025 managed to transition into full-scale industrial operation. The vast majority – approximately 90% – either remain in the testing phase, are being re-evaluated, or are being shut down entirely.

Market participants confirm these findings. MWS AI noted that this figure is consistent with the current stage of generative AI development, adding that a significant portion of initiatives are still in active development and hold potential for future scaling.

The survey revealed that 30-40% of pilot projects were halted due to the absence of expected economic returns. A key issue was the insufficient integration of solutions into companies' actual business processes. In many cases, the implemented models were not linked to core corporate systems such as CRM, ERP, or EDMS. Furthermore, some projects were initially launched more for image-building and public effect rather than practical utility, which also reduced their chances of successful implementation.

Specific case studies cited by participants in the research demonstrate typical errors. For instance, in one project, a company attempted to independently fine-tune a Chinese language model without involving specialized experts. The data used proved to be unrepresentative, and the model exhibited a low proficiency in Russian. As a result, the accuracy of the AI assistant in the legal department did not exceed 30%, leading to the project's closure. In another instance, implementation failed due to technological limitations: the company's support service processed inquiries with document and image attachments, but the models available at the time did not provide full multimodality, rendering the implementation impossible.

Financial investments in AI development are substantial. According to FinExpertiza, Russian companies allocated over 90 billion rubles to such technologies in 2024, with average expenditures per organization around 6 million rubles. In 2025, pilot project budgets typically ranged from 5 to 15 million rubles, excluding infrastructure costs.

Despite the current difficulties, the potential of generative AI is highly valued. Experts believe it can contribute to an additional GDP growth of up to 2.5% for Russia in the future. While in 2024, the technology's contribution was estimated at only 0.07-0.15% of GDP, by 2025, the effect could reach 0.5-1 trillion rubles, and by 2035, it could grow to 2% of GDP, equivalent to tens of trillions of rubles.

Concurrently, implementation timelines are shifting significantly. Half of the companies have postponed the launch of industrial solutions from 2025 – early 2026 to a later period – the second half or end of 2026. The main reasons cited are higher-than-expected project complexity, the need for employee training, infrastructure development, and ensuring security requirements.

The level of AI penetration depends on the chosen assessment methodology. For example, in terms of tool adoption and overall company readiness, the indicator can reach 80-90%. However, when it comes to deep integration into business processes, this level currently remains at 5-10%, explained by a range of limitations – from a lack of competencies to organizational and infrastructural barriers.

Experts agree that the low proportion of implemented AI projects is a normal and expected stage of technology development. The primary reasons include the complexity of integration into real business processes, a shortage of competencies, infrastructure, and mature solutions, as well as heightened security requirements. Nevertheless, the potential of generative AI remains high: as technologies are refined, experience is accumulated, and more user-friendly tools emerge, widespread adoption is anticipated in the next few years.