How We Created a Chatbot for Telecom Customer Service

  • Date Published
  • Categories Blog, Showcase
  • Reading Time 5-Minute Read

We executed a chatbot project to test how a virtual operator chatbot can automate processes for AT&T Inc. customer support.

JetSoftPro Investigates the Opportunities

JetSoftPro R&D Labs is continuously investigating the latest technologies to find the best solutions for our clients. We executed a chatbot project to test how a virtual operator chatbot can automate processes for AT&T Inc. customer support. AT&T Inc. is an American multinational telecom company and the second largest provider of mobile telephone services in the United States. We aggregated all Q&A from their customer support websites and portals, structured a database and have created a complex solution for call center automation.

This solution includes:

  • Skype text bot
  • Skype voice bot
  • Custom automated chat apps for Android and iOS
  • Social media integration

Project limitations:

  • Tight Deadlines – 1 week

We didn’t want a simple chatbot where a user’s choice is limited by several options on buttons. The chatbot had to be able to communicate with people, understand all kinds of questions and recognize human speech.

The Result (What We Built)

We created a human-like bot that answers questions about the AT&T telecom services. It provides information on available services and pricing, and addresses of nearby company stores. The chatbot processes text messages and spoken requests.

It is available:

  1. On Skype via text and voice
  2. On specially created mobile chat apps
  • It is easy to integrate into social media services like Telegram, Skype, MS Teams, Facebook Messenger, Slack, Skype for business, Twilio, GroupMe, Kik, and Bing.

Additionally, it can be integrated into any website using a WebChat interface.

The Services and Frameworks We Used

We selected a set of tools matching our project requirements.

Any bot requires basic input and output, connection to the users, and if it is not the simplest case, it must have dialog skills.

Microsoft Bot Framework provided a perfect set of tools to accomplish these tasks. One of its key advantages is that it easily allows integrating the chatbot with many of the popular channels: Skype, Viber, Facebook Messenger, Telegram, etc. Its compatibility with .Net core was helpful as we had a tight deadline and C# engineers in our R&D team.

QnAMaker was a logical choice as it’s another Microsoft tool for chatbot development. It serves to convert FAQ URLs, manuals, structured documents for a chatbot.

Understanding natural language questions and behaving like a human being was the most essential and challenging part for us, as we wanted our product to be outstanding. We had to select between two tools: Language Understanding Intelligent Service (LUIS) by Microsoft and DialogFlow by Google. When we were building our demo bot, we liked LUIS because it’s the best match for Microsoft Bot Framework, but it was in a beta version and didn’t offer the rich functionality we required. So we decided on its rival, DialogFlow.

DialogFlow allows for building conversational interfaces. It is a development suite for analyzing and understanding the intentions of users interacting with a chatbot. It helps to handle context problems, which are some of the most difficult in chatbot development. DialogFlow contains a ready-made small talks module which gives a bot a more human-like feel. It enabled our bot to answer questions like “What’s up?”, “How are you?”, “Are you there?”, etc.

We needed our bot to recognize spoken language, so we chose the Bing Speech Recognition service. This is a Microsoft tool to recognize audio recorded from a microphone in real-time and convert it to text. It is the best tool to use for Skype where we decided to implement the first iteration of our demo bot. After the voice is converted to text, our BotService processes the request and provides answers.

The Development Processes

In order to test our bot’s ability to recognize human speech, we chose AT&T, which has a detailed FAQ with public access. The data were used to train our bot.

Then, we created the logic and backend for the chatbot using Microsoft Bot Framework and QnAMaker. We added data, tested and refined the bot’s responses, but it still was not human-like. So we connected DialogFlow to teach the bot to understand customers’ intentions and context, and answer some simple questions not related to AT&T, just to chat.

We developed custom mobile chat applications using Xamarin. Forms to make our bot available on Android and iOS, and we integrated the chatbot with Skype and our custom apps.

And during all this time, we were updating its logic and functionality. We added some advanced features, e.g. processing a customer’s location to show the nearest AT&T stores.

Some Tech Problems We Encountered and Solved

  • Sending and displaying HTML-text on mobile chat apps;
  • Connecting a chatbot to Skype is poorly documented, so we did a lot of trial-and-error;
  • Creating a correct auto scroll in mobile apps as in high-end messengers.


The AT&T chatbot efficiently performs time-consuming tasks and can reduce the costs for Customer Support Service. It provides information quickly, works 24/7, and can be configured to meet your audience on any channel where customers prefer to communicate.

Visit our website to join us in working on creating new solutions and explore the currently open career possibilities.

Let’s conquer the global market together!