When chatbots first became commercially accessible, companies big and small embraced them with open arms. “Have a robot handle easy customer service questions in seconds? Amazing!” — we thought.
The problem was, these early chatbots were less C-3PO and more an annoying barrier to an actual human. From being asked: “Can you repeat the question” 10 times over to being directed to a completely unrelated information page — customers simply don't have the patience to deal with badly made chatbots anymore.
In fact, a study by Zoom found that over half of respondents would switch to a competitor after just one or two bad customer support experiences.
But could new advances in AI technology give us the smart, emotionally intelligent, and proactive chatbots of our sci-fi dreams? Let's take a look at where chatbots go wrong and how AI can help.
Going off-script
If you've ever travelled to a foreign country to test out your language skills, you'll know that what they teach you in class is completely different from how people actually speak in practice. “How are you?” may be replaced by “howzit?” “10 pounds” becomes “10 quid.” It's not until you've spent time around locals that you really learn how to speak a language.
Early bots were a lot like new language learners. Their knowledge of human language was limited to a preloaded set of questions and responses. Forget about slang or nuance, even saying “hi” instead of “hello” could throw them off. Ask them something outside their programming, and you could expect the infamous reply: “Sorry, I don't understand.”
Natural Language Processing (NLP) enables your chatbots to level up their human language skills. Rather than relying on pre-set questions and answers, NLP-based chatbots break down a customer's query into parts and analyse it for context and meaning.
This means customers can speak to these advanced chatbots just as they would a real customer service rep and receive amazingly non-robotic answers in return. ChatGPT is a good example of an AI tool that leverages NLP to better understand users' queries.
On top of that, the more NLP chatbots interact with customers, the more they learn. This means that over time they're able to provide more accurate and relevant responses based on past interactions.
Enhanced communication
So, AI-enhanced chatbots can type the talk. But can they speak the language?
Voice recognition and speech-to-text conversion are truly putting the ‘chat' in chatbot. Go back as little as five years, and anyone with even just a hint of an accent would struggle to get a response from a voice assistant. Today, using Natural Language Understanding (NLU), modern chatbots can detect languages and accents, respond in the same language, and convert spoken word into written responses using speech-to-text functionality.
This is also handy for customer service agents who want to generate summaries of their conversations for record-keeping and training purposes.
The emotional component
The purpose of a chatbot is in the name — to chat. By definition, they should be conversational. But chatting isn't just about words — it's about understanding emotion and nuance.
Humans don't always say what they mean; body language, tone of voice, facial expression, and inflection can all indicate a message that can't be captured by words alone. Which makes it all the more difficult for chatbots to understand what we actually mean.
Through machine learning techniques, modern chatbots can be trained to recognise the underlying intent behind messages. This is referred to as sentiment analysis, which allows AI models to detect whether human language has a positive, negative, or neutral sentiment behind it.
Because we're only human, we tend to use emotive language, even if we communicate with bots.
Sentiment analysis tools can grade data on a scale of how positive or negative it is, based on the language used. For example, even the best NLP tech may not be able to understand sarcasm, but sentiment analysis can be used to detect when a customer may be fuming. This technology can be used in a wide range of instances from aiding in risk analysis to detecting and alerting agents to bereavement cases.
This comes in handy for customer service teams that need to categorise and prioritise cases quickly or pinpoint which ones need to be rerouted or escalated to a human representative. This kind of intelligent routing and escalation can reduce response rates and save customer service teams time trying to match cases to the right agents.
Learning and drawing insights for the future
Common sense is one inherent trait (that most humans have), which sets us apart from our increasingly smart machines. If we do something enough times without getting the desired result, it's the little voice that tells us: “Hey, maybe something's not quite right here.”
While we still haven't been able to program common sense into our machines, predictive analytics can help bots learn from past data and provide proactive support
If a customer publishes a product review online and mentions a product fault, predictive analytics tools can help you track down customers using the same product that could face similar issues. Here's the clever bit: you can use that data to provide targeted support for affected customers, issue a mass statement about the fault, and influence future product development.
Predictive analytics might help you secure a sneaky little upsell too. By analysing customers' past shopping data, predictive analytics tools can make personalised product recommendations for individual customers.
Scaling Success: How Generative AI is Revolutionising Customer Experience (CX)
If you're ready to dive deeper into the world of AI for customer service, check out this on-demand webinar where experts Tim Banting of Omdia and Iqbal Javaid from Zoom discuss:
- Adoption trends and the most popular AI technologies right now
- Some of the challenges when it comes to data, security, and bias
- Best practices in integrating AI tools into customer service teams
- Zoom's AI-based customer experience platform