Chatbots: The Good, the Bad and the Future

Many sites offer chatbots to help visitors get quick answers and find resources. For many customers, employees and brands, they’re an ideal solution. Customers can self-serve, freeing up the bandwidth of employees who would otherwise be busy providing real-time assistance for straightforward questions. This enables customer care reps to have more time to help customers with more complex or intensive interactions. For brands, chatbots can increase CSAT, a key metric for maintaining a relationship with that customer.

But it’s foolish to deny that for many users, chatbots are extremely frustrating and can lead to a poor customer experience (CX). When agitated customers escalate an issue to a live representative, they’re likely to bring that frustration (along with the confusion that led them to seek support in the first place) to the conversation with the rep. Chances are that the employee will get an earful, resulting in a poor employee experience (EX).

What’s going wrong, and how can we fix it?

Understanding chatbots and how they function

These poor customer and employee experiences occur when chatbots are used as a panacea for all customer issues, when they shouldn’t be. Chatbots as we know them are tools designed to solve very specific questions: how to renew a subscription, how to change personal preferences or how to get a refund.

Chatbots are often programmed to follow discrete journeys—aka “happy paths”—that are planned by people who are intimately familiar with a brand’s internal workflows. These workflows are traditionally sketched out on a white board in a planning session, and the output of that brainstorm is then sent to the engineering team to program.

Chatbots work flawlessly when a customer asks a question within the scope of these workflows and uses keywords the chatbot is trained to identify. But they’re utterly exasperating when a customer has another question, or describes their issue using language or terms the chatbot hasn’t been trained on. That frustration is compounded when such bots are the only support available.

Getting to the next level of personalized CX

Sometimes, customers need an empathetic interaction. They need to feel heard. As people, we can all tell the difference between someone who is listening in order to speak next, and someone who is listening to continue the conversation. The former will pick up on a single word or topic and run with it, while the latter will identify the intent that’s actually driving the conversation.

The rules-driven chatbot is the former—the person listening in order to speak next. Is it possible to make a chatbot that is capable of listening to intent, and responding in an empathetic manner? I think yes, given the advances in AI.

Purpose-built AI for exceptional CX

AI lives and dies on the data on which it is trained. The more the data relates to the task at hand—in this case, the customer experience—the better it can produce models that accurately predict the most effective workflow for a positive and rewarding outcome for the customer.

Let’s say you’re a telecom brand, and whenever your internet service goes out, however brief the outage, your call center is flooded with affected customers. These people want to speak to a customer care person, but your call center doesn’t have enough people to facilitate this spike in call volume. Can you automate it in a way that is both empathetic and quick?

The answer is yes, as long as you can leverage AI that is purpose-built for the next generation of CX. Such a purpose-built solution consists of a vast dataset of CX interactions, along with robust analyses (i.e., models) that understands how to predict the best outcome for each CX interaction. In this scenario, the appropriate model—a service outage model—can analyze all past interactions between customers, agents and bots to predict the best processes for automated results to optimize the customer experience. Like many AI models, these types of CX models can be tuned for specific priorities, such as empathy, efficiency, ROI or scale.

This data, combined with the priorities of the brand, can be used to automate workflows, so that when an issue arises, the brand’s customers are presented with journeys that have been optimized to address specific issues and result in positive outcomes.

With the optimized journey now memorialized in a CX workflow, the next step is to enable a chatbot to serve as the go-between so that the customer can have a human-like interaction with a chatbot.

The last mile: generative AI

This is where generative AI solutions, fine-tuned with vast amounts of CX data, come into play. Large language models, such as ChatGPT, excel at human conversations, but they need substantial help in order to specialize in specific use cases, such as CX interactions. That work includes fine tuning, a process of providing the model with a vast number of inputs and outputs so it will respond to customer interactions. It also requires prompt engineering, which teaches the model with explicit examples of what the responses should be, and rigorous testing.

Granted, fine tuning and prompt engineering require massive investments and access to a database that contains CX interactions across all channels. But the end result easily justifies the investment. This work helps the chatbot better serve as an intermediary by enabling it to engage in more natural and engaging conversations, such as understanding and responding to a wide range of user inputs. The result is a conversation that’s dynamic and personalized, and sets the stage for an empathetic CX.

This capability allows AI-fueled chatbots to look beyond keywords in order to decipher intent. What is the user really asking? Are they at wit’s end? Are they suggesting a product enhancement? Does the generative AI chatbot need more information from them in order to fully resolve the issue? If yes, which specific data points or information should it request from the user?

It’s unlikely that individual brands have access to the volume of CX data and AI expertise required to train a chatbot, but they don’t need to. The providers of CX platforms are best suited to do this work, as they can see interactions across all sectors and channels.

The future is now

Hyper-focused CX data and AI can transform the role of chatbots in customer support, offering a more personalized and empathetic conversation than the traditional rules-driven approach. While chatbots still have their place in specific workflows, the ability of AI-fueled bots to understand intent and provide dynamic, engaging conversations holds great promise for enhancing the customer experience. As this technology continues to advance, organizations should prepare for the future and explore how generative AI can best fit into their customer support strategies.