Over 12,000 companies in 70-plus countries rely on HappyFox—a practical help desk and customer support software as-a-service (SaaS) platform—to optimize their customer support processes. To streamline support operations, HappyFox offers a suite of AI-powered solutions. These include integrated chatbots for seamless assistance across various channels, workflow automation with AI capabilities activated through chat features, and a business intelligence tool. The business also offers Assist AI, a Slack- and Teams-integrated chatbot aimed at increasing employee productivity.

HappyFox was using an external AI service to support its AI-powered large language models (LLMs). However, over time, the business faced several challenges with its LLM capabilities. Precision and accuracy were crucial in customer service, and the solution sometimes struggled to consistently provide factually correct responses, impacting the reliability of customer support.

Pradeek J, chief technology officer (CTO) at HappyFox, says, “In multi-turn interactions, such as with AI chatbots, the model had difficulty retaining context, leading to irrelevant or incorrect responses based on past interactions. We also encountered difficulties in ensuring that the LLM adhered to the boundaries of the customers’ knowledge base and articles and refrained from fabricating additional information."

Additionally, performance and scalability issues were evident as AI usage increased. Latency problems, timeouts, and challenges in handling complex instructions prompted HappyFox to explore an alternative AI solution.

HappyFox, whose entire production workload was already running on Amazon Web Services (AWS), chose to enhance its AI-powered solutions and introduce generative AI capabilities using Claude, an LLM from Anthropic.

HappyFox deployed Claude 2.1 on Amazon Bedrock, a managed service offering secure cloud access to foundation models. Pradeek says, “We avoided sharing data with an external vendor by implementing Claude 2.1 on Amazon Bedrock. This decision also offers a significant compliance advantage as all data is hosted within AWS.” Additionally, HappyFox no longer experiences latency and timeout challenges. Previously, response times ranged from 15‒20 seconds, but with the adoption of streaming support, response times are now consistently under 10 seconds, further enhancing the customer support experience.

Compared to its previous external AI solution, HappyFox has noted better performance and accuracy with Claude on Bedrock. It can follow complex instructions more effectively, stay within the boundaries of knowledge base articles, and provide relevant and contextually appropriate answers to chatbot queries. The solution is also more cost-effective compared to alternatives, making it an economical choice for the business and providing value for money.

Since implementing Claude 2.0 on Amazon Bedrock and transitioning to Claude 2.1, HappyFox has introduced new generative AI features and capabilities, leading to improved productivity and an enhanced customer experience.

One notable feature is its AI Agent Copilot solution, seamlessly integrated within support ticket and case pages to swiftly analyze the entire ticket for support agents. Unlike previous functionality that focused on specific actions, this feature helps the AI understand the nuances of questions. For instance, it can assist agents in composing replies by searching the knowledge base, analyzing messages, reviewing conversation history, or consulting a canned response repository to deliver accurate and comprehensive responses.

Pradeek explains, “Manually reviewing a support ticket to grasp the context takes time. With AI Agent Copilot, a support agent can request for a quick summary which includes the customer's problem, successful or unsuccessful attempts at resolution, and any pending action items.” Since introducing this feature, HappyFox’s customers have seen a 30 percent increase in agent productivity.

HappyFox is also simplifying the customer experience by developing new automated Q&A features. Now, instead of navigating through a lengthy list of questions before reaching a live agent, HappyFox customers receive prompt, automated answers. “By using Claude on Amazon Bedrock, we’ve seen a 40 percent increase in automated resolution of support tickets and chats for our help desk, chatbot, and Assist AI products,” says Pradeek.

Furthermore, the business is empowering agents to provide accurate responses, even in languages where they may not be proficient. “The AI steps in to refine responses, identifying mistakes and addressing any issues with the content before it’s delivered,” Pradeek explains. “This feature has proven to be popular among many of our early customers.”

Looking ahead, HappyFox is actively evaluating new Amazon Bedrock features, including a batch API designed for aggregating and analyzing responses simultaneously. Says Pradeek, "We have a lot of existing support tickets to feed into an LLM for data insights. The batch API feature will streamline this process, further bolstering productivity for our engineers. We’re excited to continue using Claude on Amazon Bedrock to move our business forward.”