In customer service and tech support, people often stress the importance of “the human touch.” Yet, many of us have experienced the frustration of reaching out to a software publisher for support, only to be thwarted by a clunky chatbot offering irrelevant suggestions, no matter how many times you type or shout “Let me talk to a human!”  

However, as more advanced generative AI agents begin fielding support calls, companies and customers are reevaluating the merits of automated support, and whether today’s technology is capable of giving users a better experience.  In this blog, we’ll look at when users might be happier talking to robots, and where the human touch still matters.

The Problem with Traditional Chatbots

Customer service and tech support are like insurance or medical care: something we don’t necessarily want to use, but when we need it, we want it to be fast and convenient.  

A study by Zendesk and Nuance found that 67% of users regarded calling support as an inconvenience, preferring to find their own answers via a knowledge base or Google search. So, by the time a user reaches out to support, they’re probably exhausted and frustrated from searching for answers on their own, and want a quick resolution.

This is where old-fashioned chatbots – which could only offer a limited range of scripted answers, based on keywords – fell short.  A Forrester study found 75% of customers thought traditional chatbots were incapable of solving complex problems, and a UJET survey from 2022 found over 72% of respondents declared traditional chatbot interactions a “complete waste of time”, with only a 37% resolution rate.  Unsurprisingly, 80% of respondents said interactions with traditional chatbots increased—rather than decreased—their frustration.

So, if the previous generation of automated support had such an abysmal track record, why should we expect the new generation of AI support agents to do any better?

The Difference with Generative AI

The major difference between new generative AI support agents and old-fashioned chatbots is their ability to process language. Rule-based chatbots simply scanned the user’s input for keywords, then pulled up any knowledge base content tagged with the same keyword.

For instance, if a user typed, “My software keeps crashing when I try to save my work,” a rule-based chatbot might reply, “Here are some tips on how to save your work,” completely missing the issue of the software crashing. In fact, 50% of users in the Forrester survey said that traditional chatbots regularly gave answers that made no sense in the context of their issue.

By contrast, generative AI can process not just keywords but grammar, sentence structure, and other intricacies of language, allowing modern AI agents to grasp the full context of a user’s issue and pull the most relevant information from the product documentation. Beyond that, generative AI agents are capable of reasoning, allowing them to ask follow-up questions and troubleshoot in the same way a human support rep would.

Presented with the same complaint—”My software keeps crashing when I try to save my work”—a generative AI agent might reply, “Sorry to hear that. Does it happen every time you try to save your work? Is it just that particular project? Have you tried creating a new project and saving it?”

While the technology is still too new to gather comprehensive data, early pilots of generative AI support agents have been promising.  The Swedish fintech company Klarna deployed a ChatGPT based support agent that now handles two thirds of its live chat interactions (over 2 million conversations to date).  In a short period, the AI agent reduced average call time from 11 minutes to 2 minutes, with a 25% improvement in first-time resolution rate.  Similarly, Clickup, a project management software provider, saw a 20% improvement in automated resolutions after upgrading from a scripted chatbot to an AI agent.

New Possibilities

While AI agents outperform traditional chatbots in nearly every important service metric, their true potential lies in delivering entirely new forms of customer support.

For example, our company’s app, Parrotbox.ai, allows generative AI agents to read the text on a user’s screen via a browser extension, enabling the agent to guide users through tasks and troubleshooting steps live, just like a human support rep might via screen sharing. This capability lets software providers deliver highly tailored support instantly, at the point of use, before users resort to Googling or visiting a support portal.

And because generative AI agents can understand the larger context of customer questions, they can provide in-depth guidance on how users can leverage a software app to solve specific, real-world problems.

In addition to providing user support directly, generative AI agents can also collaborate with human support reps, providing fast answers to product knowledge questions as well as personalized training and coaching.  A study by Stanford University found that call center agents paired with AI agents increased productivity by 14% (up to 34% for new hires.) 

Open the player for full screen

Conclusion

Generative AI agents are not just a step above traditional chatbots—they’re an evolutionary leap. While the human touch remains invaluable in certain contexts—such as customer success conversations, upselling, and addressing serious complaints—the efficiency and instant availability of AI agents make them equally or more effective than human reps for most support interactions.