Tiqum analysis: AI assistants sped up customer service 1.5 times in 2025
The digital integrator Tiqum presented data on the effectiveness of LLM deployments in big business: how the RAG approach solves the problem of AI “hallucinations”

A breakthrough in the speed of handling requests
The main achievement of the year was the introduction of intelligent assistants in customer support and internal HR services. According to Tiqum's research, the use of large language models (LLMs) let companies cut the time spent handling initial customer requests by 40–50% on average.
The key success factor was the move from “isolated” chatbots to systems using RAG (retrieval-augmented generation). This approach lets an AI assistant consult the company's private knowledge base in real time. The RAG architecture made it possible to:
- Reduce the number of errors: the accuracy of answers to complex technical questions rose to 94%.
- Minimize “hallucinations”: the AI no longer invents the terms of promotions or tariffs and relies only on facts from the company's documents.
- Automate routine: up to 70% of standard requests are now closed without an operator.
The economic effect and the barriers
The introduction of AI assistants in 2025 shows one of the highest ROI figures of all IT initiatives. In companies with more than 500 employees, the savings from optimizing payroll and speeding up support processes start at 15 million rubles a year.
But Tiqum's research also revealed a significant barrier: 45% of the companies that tried to adopt AI on their own ran into problems integrating the models into their existing IT landscape (CRM, ERP).
“AI is not a magic button but a complex layer on top that requires impeccably clean data and a clear architecture of connections,” the company's experts stress.
The forecast for 2026
Tiqum's analysts predict that next year the trend will shift toward multimodal assistants able to work not only with text but also with voice, photos of documents and video streams in real time. Companies that have already built a knowledge base for AI will get a multiple advantage in speed of service and lower operating costs.


