> 95%

Automated Query Resolution

~30%

Reduced in Cost

In this case study

Gaming

Industry

Overview

Client overview

Our client is a leading blockchain gaming company with 20M+ gamers in its network. They have users in 20+ countries and receive around 10k+ monthly chatbot queries. 

Due to the high volume of queries, they were unable to provide quick responses to the customers, especially in non-office or rush hours (example when new events and campaigns were launched).

Challenge

The customer support challenge

Our client was using a button based flow over their chatbot to address customer queries. The customers had to do step-by-step navigation to get their answers. Due to this lengthy multi-step process, the chatbot had become inefficient at resolving customer queries. Hence, 75%+ chats were handed over to human agents for resolution. 

High rate of human handover led to a high average resolution time and also increased customer support costs significantly. To solve for this, our client wanted an AI powered chatbot that provided quick, cost-effective resolution.

Specifically, they wanted the chatbot to have the following features: 

  1. Ability to understand free flowing language and also engage in human-like conversations
  2. Generate personalized responses based on the current context as well as any past interactions with the chatbot
  3. Could integrate with seamlessly with multiple channels like Chatbot, Telegram and Discord
Solution

Robylon's solution

There were 2 major categories of queries our client received : Customer specific queries (eg payment issues, tech issues) and General Knowledge Base (KB) queries. 

We realized that 60%+ queries were related to KB wherein the customers wanted to know specific details around upcoming campaigns, events, rewards T&Cs, etc. Such queries were largely repetitive and could be easily automated using our AI agents. 

Our approach is briefly described below: 

Step 1: Triage agent for intent identification

A triage agent was made for customer intent identification (eg, identifying if a customer query is account specific or KB related). For KB queries, the chat was handed over to our Knowledge Base agent. 

For accurate intent identification, the agent was trained on 25k+ historical chats. This mapped the user’s intent with a list of keywords/ phrases they actually used. This mapping helped achieve over98% accuracy in intent identification. 

Step 2: Knowledge Base agent

A specialized knowledge base (KB)agent was made for blockchain-based gaming companies. It was trained with industry-specific knowledge to develop domain expertise. On top of this, client-specific information was also provided to develop context on the client. 

We solved 2 specific challenges in answering the queries:

  1. Accuracy - Our custom designed flow captures user feedback on the quality of the answers and incorporates it back in the AI agent. This helped us moved the accuracy from ~85% to over 97% in a little less than a week.
  2. Quick Update - Our client keeps on launching new campaigns every week. Each campaign has an associated set of newFAQs. Robylon's 'Custom Q&A' agent helps our client update each such change in less than 10 mins.

Step 3: Responding agent

The responding agent is designed to engage in human-like conversations with the customer. It takes input from the Knowledge base agent and the conversation history. It then generates personalized responses for the customer. 

Additionally, the agent is capable of conversing in 40+ languages (English, French, German, etc.) to address customer queries across 20+ countries.

Step 4: Escalation to human agents

Chats where the AI agent is unable to resolve queries were escalated for human intervention. This ensures that the customer is not stuck in a loop dealing with an AI chatbot. Additionally, for all the conversations, thecontext is readily summarized and provided to the human agent during the handover. 

Result

Out of 6k+ Knowledge base queries monthly, Robylon’s chatbot was able to resolve 95% queries automatically. Only ~10% KB chats were escalated for human intervention. This not only reduced the average resolution time but also reduced customer support costs by 30%. 

Our client also saw significant improvement in their CSAT scores as our chatbot provided quick and accurate replies which were also highly-personalized.

Other case studies