RChee - Customer Support at Scale
Customer support is an integral process in any organization, and it's no surprise that it is estimated to be a $350 billion-a-year industry. More and more businesses today are investing in Conversational AI and integrating it into their CX (Customer Experience) strategy. And for good reason. In fact, thirty-four percent of sales and marketing leaders believe that artificial intelligence will cause the most sizable improvement in customer experience.
RingCentral's Rchee chatbot is a great example of the effective use of Conversational AI that assists human support agents, instead of replacing them. The agent has over 2,000 intents configured and it is capable of answering any frequent customer queries and FAQs without the intervention of human support, resulting in hundreds of hours of agent cost savings for the company. It can easily resolve a high volume of simple tasks: help clients to change passwords, create cases, get answers to certain questions, and schedule appointments in a way that mimics natural non-linear conversation.
Some of its other intelligent features include -
⦁ Identify users and route them to right support queue based on their account.
⦁ Gather certain required information before handing a conversation to human agent.
⦁ Expand business hours: gather leads and provide support even when no human agents are available.
⦁ Image Optical Character Recognition.
⦁ Smart default fallback behaviour.
That last one is really important. The default fallback behaviour of a chatbot is crucial and it is the response that is triggered when the bot does not understand user's input. This happens more often than you'd expect. Most chatbots have a generic default message like "I'm sorry I didn't understand that. Can you say that again?" but this is far from ideal and quite honestly really frustrating as it can get the user stuck in a loop.
Gracefully handling fallbacks is an important behaviour to be considered when designing your conversational assistant. Rchee has a great dynamic fallback behaviour that takes the user's input, passes it through Coveo Search (internal document search engine) and returns support articles relevant to the user's query. This is quite effective as it addresses the user's concern and shows them potential solutions, even though the bot does not understand the user's intent.
My Role :
As the lead chatbot developer, my role was mainly to manage the operation & behaviour of the virtual conversational agent. This includes writing webhook fulfillment code to build new features or to modify/update the existing intents, analyzing customer chat history to observe any inconsistencies, and constantly updating the intents, making the bot smarter in every iteration.
My Contributions :
1. MegaAgent Migration - A single agent in Dialogflow is limited to 2,000 intents. But due to the large scope of Rchee, it needed more than 2,000 intents which requires a Mega agent in Dialogflow to increase the intent limit. I was solely responsible for migrating Rchee to a MegaAgent within Dialogflow and Google Cloud Platform.
This was no simple task as the entire codebase needed to be modified to support the MegaAgent. I was able to successfully migrate the existing project and upgrade it to a MegaAgent, enabling it to now support a maximum of 20,000 intents, freeing the team more than enough space to add newer intents and expand the bot's intelligence.
2. Live Agent Deflection - I was responsible for developing the 'live agent deflection' feature which keeps the user from reaching the support agent directly, and prompts them to enter more details before handing them off to the support queue. Most chatbots directly handoff to an agent when the user asks for a live agent, but this is not advised as it can burn a hole in your pocket. Almost 90% of users' queries are pretty simple and can be handled by the virtual agent alone, and it's essential to collect more information to check if the bot alone can resolve their query first.
Adding this extra step before handing over to human support is a small but powerful trick you can use to increase the efficiency of your large-scale chatbot. This critical feature reduced live agent handoff rate by more than 70%, adding a ton more hours of extra cost savings for the company.
3. Intelligent Image OCR - I was also responsible for developing Image Optical Character Recognition functionality in Rchee. The feature allows users to upload a photo or screenshot of the issue they're facing and the bot extracts text from the image, passes it over to Coveo search and returns matching support articles. This is surprisingly effective and returns relevant support documents with a high accuracy, making it easy for users to troubleshoot any issues they may be facing.
In conclusion, the customer service industry is rapidly being disrupted with the advent of Conversational AI and chatbots. As these bots become smarter, they will soon be able to understand and help us humans better and will predictably play a larger role in our everyday lives.