Case Study: Fourthwall
"Creators are extremely weary of AI and have little patience for bots. Decagon has delivered an AI solution that has exceeded our expectations and those of our content creators, particularly by being very human-like and conversational. We are now fully resolving over 70% of incoming tickets, and Decagon's solution is effectively a high-functioning agent that gives personalized responses instantly and is always online. We're excited to continue working with them as we scale our business."
Eli Valentin
COO at Fourthwall
Overview
In today's creator-heavy world, Fourthwall helps content creators (e.g. Youtube stars, Twitch streamers, etc) consolidate their various income streams within one all-in-one creator platform, aiming to take away some of the backend administrative stress of running your own business. Fourthwall offers a white-label website with built-in tools for setting up merch stores, membership programs, accepting donations, and more, with over 200 different types of products that creators can develop to sell through their website.
Content creators are not an easy audience to please, often skeptical of new products and demanding of high-quality customer support. Fourthwall's team was looking for a way to scale their customer support operation without sacrificing the quality of the customer experience. To accomplish this, it was critical to provide a very conversational experience, and not the traditional chatbot with canned responses. It absolutely had to feel like you were talking to a human. They were also looking for a way to leverage the efficiencies of AI in their support operation, which is manned by a team of human agents.
Content creators are not an easy audience to please, often skeptical of new products and demanding of high-quality customer support. Fourthwall's team was looking for a way to scale their customer support operation without sacrificing the quality of the customer experience. To accomplish this, it was critical to provide a very conversational experience, and not the traditional chatbot with canned responses. It absolutely had to feel like you were talking to a human. They were also looking for a way to leverage the efficiencies of AI in their support operation, which is manned by a team of human agents.
The Problem
Standard customer support workflows weren't providing creators with the ideal experience. Prior to Decagon, new tickets were largely handled manually by agents. Tickets were often in the form of short, targeted questions and feedback messages from creators, who often had a high expectation of response time and quality.
The team needed a reliable way to give creators near-instant answers around the clock, as the user base is global and growing rapidly. Moreover, it was especially important to not give creators an "oldschool" cookie-cutter chatbot experience that simply regurgitated canned responses and Help Center articles -- nothing would frustrate creators more. It had to be a very human-like experience.
Finally, being a very tech-driven company, the team was acutely aware of the cost efficiencies a well-implemented AI solution could bring in being the first line of defense. Not only are a majority of tickets automatically resolved, but the agents are now freed up to do more high-value work.
The team needed a reliable way to give creators near-instant answers around the clock, as the user base is global and growing rapidly. Moreover, it was especially important to not give creators an "oldschool" cookie-cutter chatbot experience that simply regurgitated canned responses and Help Center articles -- nothing would frustrate creators more. It had to be a very human-like experience.
Finally, being a very tech-driven company, the team was acutely aware of the cost efficiencies a well-implemented AI solution could bring in being the first line of defense. Not only are a majority of tickets automatically resolved, but the agents are now freed up to do more high-value work.
The Solution
Because of the direct nature of many of the tickets (e.g. "How do I do X?", "Does your product support Y?", "I'm having trouble with Z"), Decagon's fully generative AI solution was effective from the start. Decagon is able to reference relevant internal info about the product and user, draft a personalized response to any question, and resolve the ticket autonomously. This is done by leveraging the latest techniques in generative AI, including Large Language Models (like ChatGPT) and vector embeddings.
The solution consisted first of ingesting the complete knowledge base of the product, including all Help Center articles, FAQs, and other relevant documentation, as well as learning from historical tickets handled by the top agents. This was a seamless out-of-the-box experience, and required no engineering lift from the Fourthwall team. Decagon and Fourthwall also worked together to feed in relevant information about the user, such as their shop URL on Fourthwall. This information is annotated as metadata on the conversation and is used by the AI for personalization and basic decision-making.
In cases where the AI is unable to fully resolve the ticket, it is escalated to a human agent. This is done in a way that is seamless to the user, and the agent is able to pick up the conversation where the AI left off, which is key since everything can flow to and from Fourthwall's existing ticketing system.
A key differentiator in the final solution was the ability for Fourthwall leadership to have full visibility into the AI's performance through the admin portal, and to be able to tweak and tune the AI's behavior in real-time. Metrics and insights are computed automatically in real-time, something that would've taken human agents days to manually maintain (e.g. tagging every conversation to see what users are asking at a high level).
The solution consisted first of ingesting the complete knowledge base of the product, including all Help Center articles, FAQs, and other relevant documentation, as well as learning from historical tickets handled by the top agents. This was a seamless out-of-the-box experience, and required no engineering lift from the Fourthwall team. Decagon and Fourthwall also worked together to feed in relevant information about the user, such as their shop URL on Fourthwall. This information is annotated as metadata on the conversation and is used by the AI for personalization and basic decision-making.
In cases where the AI is unable to fully resolve the ticket, it is escalated to a human agent. This is done in a way that is seamless to the user, and the agent is able to pick up the conversation where the AI left off, which is key since everything can flow to and from Fourthwall's existing ticketing system.
A key differentiator in the final solution was the ability for Fourthwall leadership to have full visibility into the AI's performance through the admin portal, and to be able to tweak and tune the AI's behavior in real-time. Metrics and insights are computed automatically in real-time, something that would've taken human agents days to manually maintain (e.g. tagging every conversation to see what users are asking at a high level).
The Result
Ultimately, Decagon delivered a solution that yielded CSAT and resolution rates that far exceeded expectations, resolving over 70% of tickets without a need for human eyes at all. Furthermore, the solution was very well received by the creator community (not an easy audience to please) and delivered the superior user experience that Fourthwall was looking for. Furthermore, the Decagon team proved to be very capable in swiftly handling any feature requests that came up from both Fourthwall and the creator community.
© 2024 Decagon. All rights reserved.
