Freshworks, the NASDAQ-listed SaaS giant from Chennai, is integrating “AI as the new UI” across its platform. With their Freddy AI co-pilot, “businesses can now transform customer service and support operations from being just a cost center to really become a revenue center and drive future sales,” says Priya Subramani, Vice President and General Manager for customer experience products at Freshworks.
Freshworks, for those who do not know already, began its journey in 2010 with Freshdesk, a ticketing system designed to streamline customer support. Since then, it has expanded its product portfolio to include Freshcaller telephony solution, Freshchat messaging platform, and more. As of 2024, Freshworks offers a suite of products for customer service, IT management, and CRM, helping over 67,000 clients in 120+ countries to connect to their customers.
As a CX product leader at Freshworks, Priya believes, “Delivering exceptional CX is all about how you use every instance of customer interaction—in sales, support, and more—to not just resolve issues and answer queries, but also to build your brand.”
True to this principle, Freshworks’ AI co-pilot Freddy empowers businesses to not only resolve customer queries but also deliver brand value by decoding customer intent across multiple touchpoints and interactions.
“Freddy for us is all things AI, and AI has become the new UI. Freddy is in all our products, and we want to bring its intelligence everywhere, from the bots that guide the agent to intelligent insights that help the CX leader identify problems and find effective solutions or enhancements. All of that is powered by Freddy,” says Priya.
Origins of Freddy in Freshworks: More than Just a Response to AI Trends
“For example, the early ML version of Freddy could tell you that 50% of your queries revolve around a specific theme," says Priya. "This allows you to automate responses or create workflows that address these common issues, thereby maximizing efficiency."
“If you think about the customer experience world, one of the main metrics is to reduce customer effort. So, the customer effort score (CES) is a big part of it. It’s about how easily you provide the customer with the answer. That was the idea behind Freddy when we started. Today, GenAI has allowed us to push the boundaries and explore new possibilities.”
The Three Facets of Freddy
Broadly speaking, Freddy is used to cater to the needs of three different personas. To start with, Freddy’s chatbots—their autonomous agents—are used directly by the end users. “It’s purely self-service. The end user can ask any question, and the bots can process the answer, resolve problems, or do the required tasks, all without any human intervention,” explains Priya.
The second part of Freddy is the co-pilot, which guides the agent. When a customer interacts with the customer service agent, the co-pilot guides the agent about the next steps, processes, information retrieval, and summarization. “It can even fix the tone, rephrase sentences, pull up the right information, and do anything that the agent might do,” Priya adds.
The most impressive aspect is Freddy’s ability to handle multi-turn conversations and come up with answers using GenAI without the need for scripted responses.
The third facet of Freddy AI is very interesting. Priya dubs it Freddy’s “Insights”. The insights are beneficial to C-level executives, supervisors, and administrators to understand process flows, roadblocks, and to make efficient workflow automations for the agent and the organization.
For instance, if Freddy detects a spike in average handling time in a specific region, it can pinpoint the root cause and recommend targeted actions, or “insights”.
This is made possible by sophisticated features like query-to-graph, where users can simply ask, “Show me the day with the highest average handling time,” and get precise insights on why this is happening—whether it’s due to new product queries, pricing concerns, bugs, or other factors—without sifting through complex dashboards.
In the past, building such interactions required painstaking manual scripting, but with GenAI, Freddy can now understand context, maintain conversation history, and offer intelligent responses, all in real-time.
This level of detail is a game-changer for supervisors, administrators, or C-level executives, enabling them to make data-driven decisions with ease.
Quality Coach and Thank You Detector: Small Innovations with Big Impact
As AI and GenAI continue to advance, they unlock new possibilities for solving users’ problems that were previously thought to be unsolvable. Case-in-point are two of Freddy’s most-used features: Quality Coach and Thank You Detector.
The Thank You Detector is a deceptively simple feature that solves a common problem in customer support: the reopening of closed tickets due to a customer’s follow-up message, which often is just a “thank you.”
By intelligently detecting the context of the message, Freddy prevents unnecessary ticket reopenings, saving time and reducing clutter in the system. “It might seem trivial, but the Thank You Detector has been one of our most popular features,” Priya shared, highlighting the cumulative time savings and efficiency gains it provides.
The Quality Coach, on the other hand, leverages AI to monitor every interaction between agents and customers in real-time, providing instant feedback to ensure the highest quality of service. Traditionally, companies could only audit a small sample of customer interactions, typically around 2%, to assess the quality of service. However, with Freddy, every interaction can be automatically evaluated against predefined criteria.
"Quality Coach can analyze whether the agent greeted the customer properly, whether the tone was appropriate, and whether the right information was provided," Priya explains. "It doesn't just stop at identifying issues; it also provides personalized coaching for each agent, highlighting areas where they can improve."
How does Freshworks understand user problems and develop AI products?
Priya says, “One of our biggest advantages is that we have tonnes of data. We have over 60,000 customers on the CX side, and we leverage the wealth of data generated. This data allows us to look at various themes and problems in a particular industry or service vertical for which people reach out for support repeatedly. This allows us to build the right kind of models.”
Freshworks’ own internal support teams play a crucial role to help their product teams understand how their daily work looks like, understand pain points, and internally test solutions that make their lives easier.
For instance, Freshworks’ product team learned that their customer service agents spend 45 minutes every day discussing unresolved support tickets that should be prioritized and the ones that are likely to escalate. This pain point led to a feature called ‘likelihood to escalate’ that utilizes AI to prioritize tickets based on the sentiment of the customer.
Priya says, “We are always looking for ways to understand what's really happening on the ground and how we use technology to really help solve the problems. These are instances that we look at from an AI lens.”
Priya says, “Whether it’s talking to CX leaders, our own customers, analysts, or administrators who set up and configure the products, we strive to uncover opportunities to build capabilities that genuinely address customer needs.”
“We look for applicability across different industries, and the product needs to work across businesses of all sizes. The total addressable market (TAM) for the products is important here, and it should be able to make an impact on a variety of customers and scale. A lot also depends on how passionate we are about solving that particular problem statement and what's the revenue engine that will drive growth. A new product development at Freshworks is an intersection of all of these things.”
Challenges and Lessons from Building Freddy AI: Accuracy, Adoption, and Costs
Priya detailed FreshWorks' approach: "We conduct various tests, including precision and recall measurements. We also run Freddy through a private beta phase with a select group of customers before a broader launch for the general audience. In addition, we also test it with our own internal customer support teams and implement their feedback."
This meticulous process helps FreshWorks identify and address issues, ensuring Freddy’s responses are both accurate and contextually appropriate.
The next challenge was measuring the impact of Freddy. Priya highlighted the importance of adoption metrics and user feedback.
“We look at how many agents are using the features and how often. For instance, features like conversation summarization and rephrasing have been well-received by agents, significantly improving their productivity,” she noted.
FreshWorks also tracks deflection rates, which measure how often Freddy resolves issues without human intervention.
"We aim to see steady improvements in deflection rates. Some customers have achieved up to 95% deflection with Freddy, while others are still working towards this target," Priya explained.
One of the most significant challenges for any organization today is the cost of advanced AI models. Priya addressed this by discussing FreshWorks’ approach to cost management.
“We start with rough calculations of compute and latency costs and continuously optimize our implementations to manage expenses,” she said. Despite the potential for increased costs, FreshWorks remains optimistic that ongoing research and development will drive these costs down over time.
Pricing strategies are also evolving. FreshWorks has adjusted its pricing model for AI features like Freddy, balancing between making AI accessible and covering costs.
"We look at how features impact agent productivity and adjust pricing accordingly. We may increase prices as we add more features, but we also aim to make basic features available in lower plans to encourage wider adoption," Priya explained.
"I would say, just look at what's available and build on top of it. You cannot compete against the likes of OpenAI," Priya advised.
Road Ahead: Developing Autonomous Systems and Building Trust
Looking forward, Priya envisions a future where Freddy evolves into a more autonomous system. “Our goal is to move towards what we call ‘self-healing’ systems,” she notes. “Freddy will not only provide insights but also suggest and implement changes autonomously, based on evolving data and patterns.”
However, Priya acknowledges that the journey towards full autonomy will require building trust with users. “The challenge will be in getting people to trust these systems enough to let them operate autonomously,” she said.
So, every time Freddy suggests a course of action, it provides the rationale behind it, citing the sources of information. This transparency ensures that users remain in control and can validate Freddy’s recommendations.
"We don't just tell the agent what to say; we show them where the information came from, whether it's a knowledge base, a solution article, or another source," Priya explains. "This approach fosters trust and allows agents to feel confident in the AI's guidance. It makes the transition to more autonomous operations smoother."
Freddy represents more than just an AI tool for FreshWorks; it embodies the company’s commitment to pushing the boundaries of what’s possible in customer experience.
By reducing customer effort, improving agent productivity, and providing deep insights, Freddy is not only enhancing the way businesses interact with their customers but also setting a new standard for the industry.
