Bank clients contact teams through calls, Telegram, email, app messages, and direct employee chats. At the same time, support employees spend too much time searching across FAQs, product passports, instructions, and internal documents. Requests can be lost or duplicated, and answer quality depends heavily on the individual employee.

Clients expect instant digital communication, while banks need traceability, SLA control, and consistent service. LLMs and lightweight classifiers can make request intake smarter without replacing employees.

A support co-pilot plus request intake platform: AI classifies the client message, retrieves relevant information from approved knowledge sources, drafts an official response for the employee, creates a CRM ticket when needed, assigns ownership by client/product/region/urgency, tracks SLA, and feeds analytics to managers.

• Start with Telegram as the primary client channel.
• Use AI to classify request type, urgency, and responsible department.
• Add SLA reminders and escalation for overdue tickets.
• Create a unified employee CRM workspace with request history.
• Connect feedback quality metrics to employee service dashboards.
• Start with an internal support assistant that drafts verified answers for employees before exposing anything directly to clients.
• Use retrieval-augmented generation over FAQs, product documents, tariffs, process instructions, and approved templates.
• Combine answer generation with ticket creation so unresolved questions become accountable requests.

• Conversational AI or CRM workflow team
• Experience with Telegram bots or web chat interfaces
• Strong backend workflow and notification skills
• Ability to design enterprise-grade status tracking and reporting

• Client request bot
• Client identification using TIN/client code, company name, contact person, phone number, and assigned manager
• Request categories: complaint, consultation, callback, credit, deposit, FX/conversion/payments, technical problem, other
• Auto-routing by employee, product, region, urgency, and category
• Employee CRM board with statuses: new, in progress, waiting for client, closed
• Dashboards: request volume, complaints, average closure time, overdue requests, employee workload, frequent problems
• Knowledge-base ingestion and search
• AI drafted official response for support employees
• Human approval before client-facing answer

• No request enters without an owner and status.
• Average response and closure time become measurable.
• Overdue requests are visible and escalated.
• Management can see the most frequent client problems and employee workload.
• Support employees can answer frequent questions faster without leaving the assistant.
• Drafted responses cite approved internal knowledge or show the source used.

• The system should not expose sensitive client data in insecure chat channels.
• Human handoff must be simple and reliable.
• The bot should reduce friction, not force clients through a long menu.