PSB shows how it deploys AI in the bank: chatbots, RAG and services for business

PSB has revealed not just a set of AI experiments but a fairly tough, pragmatic scheme for bringing AI into the bank's operations. The bank stresses that it did not try to be among the first to test every LLM tool. Instead, it consistently built solutions where a business effect is quick to see: serving small and medium businesses (SMBs), automating consultations and reducing the load on employees. As a result, PSB already runs a stack of scripted bots, a generative assistant, RAG search over a knowledge base and pilots for internal teams.

In external services, the main focus is on entrepreneurs. PSB uses recommendation models in its CRM to analyze transactions and suggest better-suited tariffs, products and reasons to contact a client. The logic is simple: AI should not entertain but save money and time. For example, the system can notice that a client regularly overpays for operations above a limit and would be better off switching to another tariff. Or it can suggest moving a counterparty to PSB if that lowers payment costs for both sides. According to the bank, about 98% of services for entrepreneurs are already available remotely, and 74% go through end-to-end automation with no human involvement at intermediate stages.

The key product here is the AI assistant Katyusha, available in internet banking, mobile banking, on the website and in VK. It is used not only as a chatbot for answers but as a practical tool for routine business tasks: preparing sales copy, writing marketplace product-card descriptions, replying to reviews, drafting social media posts and even producing interview questions for candidates. The bank cites about four thousand requests per quarter and stresses that the service remains free for SMB clients. It runs on a Russian LLM that PSB began testing in an early pilot. The idea is to gather the most frequent scenarios in a single window instead of sending the client to various external services.

The next step is introducing RAG into Katyusha's consultation layer. According to PSB, this layer delivered the most noticeable quality jump: in the pilot, request automation rose by 7%, average resolution time fell eightfold, and customer satisfaction with the chatbot rose by 21%. At the same time, the share of incorrect answers was cut from 3% to 1%. For the bank this matters, because at a high level of automation each extra percentage point is much harder to win than at the start.

The LLM's ethical filter proved a separate problem: in tests it blocked almost 30% of requests because of wording in the knowledge base. After rewriting the texts, replacing sensitive words and adding post-processing, the bank cut the share of such blocks to 1%, which made the pilot viable.

Another track is banking in messengers and internal AI for employees. Since 2023, Katyusha has worked in VK not only as a consultant but as an interface for basic settlement and cash services (RKO) operations: issuing invoices, making payments, and viewing statements and balances. In 2025, the number of active users of this channel grew by 35%, and more than 60 thousand payments went through it. PSB then launched payment services in MAX as well.

Inside the bank, the employee pilot started in October 2025, first with 25 people who need to answer requests quickly from sales offices and field managers. The goal is to free up about 10% of working time, but the pilot showed that AI delivers the most value not on routine questions but in narrower scenarios without personal data, where precise consultations and product hints matter.

A fairly clear strategy emerges. PSB is not trying to build a universal AI bank in one step. It separates tasks by level of risk and benefit: cloud models for content and non-personalized scenarios, scripted bots for quick simple answers, RAG for scalable consultations, and an in-perimeter LLM for sensitive data and internal services. In 2026, the bank plans to deploy an LLM inside its own perimeter and raise the share of RAG answers in the overall consultation flow from 4% to 16%.

For the market, this is an instructive example of a large bank deploying generative AI not for show but as a working tool with clear metrics.