Services and platforms

AI operator for a subscription payment service

An AI assistant in Telegram answers customers in place of an operator and works out the order price itself — only complex questions go to a person

Python 3.12aiogram 3FastAPIPostgreSQL 16Redis 7arqQdrantDocker Compose
AI operator for a subscription payment service

Project at a glance

Problem

A payment service for foreign subscriptions and digital goods was receiving a growing number of requests. Staff advised buyers and handled orders by hand, and scaling customer service the same way was becoming difficult. The client needed to automate the whole path from consultation and price calculation to payment and issuing virtual card details. Non-standard questions still had to stay with live operators.

What we did
  • Recognition of colloquial service names
  • Catalog and order price calculation
  • Handoff to an operator with the reason for the request
  • Distribution of requests among staff on shift
What the client wanted
  • Automate customer service as the number of requests grows
  • Tie together consultation, pricing, payment and issuing card details
  • Pass non-standard questions to a live operator

Want the same for your business? Tell us about your task — we’ll propose a solution and estimate the timeline.

Consultation and choosing a service

We built a Telegram assistant that understands the colloquial names people use for services and shows the catalog and plan options. To price an order, it asks for the amount and the currency.

Typical questions are handled by rules, with answers taken from the client’s library. The catalog has 77 items. They are ordered by demand, determined from customer requests, so buyers see the most popular services first.

An operator joins the conversation

When a customer asks for a human, complains about a payment, or asks about a refund or a confirmation code, the assistant passes the conversation to an operator. Along with the conversation, the operator gets the reason for the request.

The admin panel shows waiting and active conversations. New requests are distributed round-robin among the staff on shift. Search helps find a conversation, and canned replies can be edited without a developer.

An operator can step into a conversation from their phone. The assistant then goes silent and automatically resumes after two hours.

Pricing and order processing

Order prices are calculated by code. Prices and exchange rates come from the calculation tools; replies in which the language model states these values on its own are blocked.

We built order state management and payment notification processing with six checks. Conversations, payments and outgoing messages run in three separate queues, each with its own workers. If the system is overloaded, the request goes to an operator.

The full virtual card number and security code are not stored in the database — only a masked number remains. The message with the card details is also kept out of the system’s conversation history.

Running the service from one panel

The client got a Telegram assistant and an admin panel for managing the catalog, service modes, the exchange rate and orders. Operators handle requests from a shared queue and use an editable library of replies.

The financial summary shows revenue, markup, and the costs of issuing cards and accepting payments. Margin calculations take both types of costs into account.

Target audience

  • Buyers of foreign subscriptions and digital goods
  • Operators of the payment service
  • Staff who manage the catalog, exchange rate and orders

What users need

Job to be done — from situation to outcome

Situation

I have a card from a Russian bank, but to pay for a foreign subscription I need foreign card details and instructions on how to use them.

Need

I want to pay for the foreign service I need and know that I’ll get help if something goes wrong.

Outcome

I can start a purchase with my Russian bank card, get foreign card details and pay for the service I need, with support along the way.

What’s inside

1

Recognition of colloquial service names

2

Catalog and order price calculation

3

Handoff to an operator with the reason for the request

4

Distribution of requests among staff on shift

5

Conversation search and editable reply templates

6

Assistant pauses when an operator steps in

7

Order management and a financial summary

8

Only a masked card number is stored, without the full number or code

How we worked

1

Analyzing demand from customer requests

2

Building the Telegram assistant

3

Creating the operator workspace

4

Building price calculation and order processing

5

Creating tools to manage the catalog and finances

Need a similar project?

Tell us about your task — we’ll estimate the scope, timeline and cost and propose a solution.