AI is weaving itself into our daily lives rather rapidly. Many businesses are using AI to automate repetitive and mundane tasks. Consequently, (conversational) AI is becoming a viable option for the customer support domain - more specifically in the form of AI-powered chatbots.
A search for ‘AI customer support chatbot’ in Google provides you with many websites telling you how they can level up your business and which are the best ones. However, the question of whether an AI chatbot is beneficial to your business case specifically does not seem to be taken into consideration. We’ll assume it is your goal to service your customers to the best of your ability as it creates a positive association with your brand. Ironically, many companies are making the switch to AI primarily as a way to save money and service customers more quickly - though arguably, not more effectively. Has artificial intelligence grown enough that we can trust it to serve our clients?
The rise of AI as an assistant
The idea of AI assisting customers is nothing new. Remember Clippy the Office Assistant? Well, so do we - and not for good reasons. With Clippy, Microsoft was trying to service customers without having to involve an employee or the customer having to search for an answer themselves. It sounded like a great idea, but Clippy is now remembered as the cause of many frustrations for being invasive and not understanding human intention.
Nevertheless, Clippy is a good example of a first step towards using AI in customer service. Bill Gates, the owner of Microsoft, once said in 1995 about AI: ‘The whole way you interact with the machine will be different, you'll be able to talk to the machine, and it will use voice recognition, or so-called natural-language processing, to be able to understand what you do.’ And he was right. Other AI tools, for example, Siri, soon followed, weaving their way through our technological to make our lives a little easier.
The real game-changer that we are seeing today, making AI interesting for customer service, is the introduction of large language models (LLMs). LLMs make recognising and generating text much easier for AI. As a result, human dialogue can be analysed and mimicked more accurately. At that point, the link to chatbots was easily made and a new type of customer service employee was born.
What are chatbots and conversational AI?
It’s safe to assume that we are all familiar with chatbots. We often encounter them while shopping online or trying to find more information on company websites. The chatbots as most of us know them are ‘rule-based chatbots’. Most notably, they can respond to input according to a strict set of predefined rules. In most cases, the user is offered a drop-down menu of options to converse with the chatbot or the bot will only respond to very specific keywords. It’s not uncommon to be met with a “Sorry, I did not understand your question” during your conversation. Rule-based chatbots are simply not open to exceptions or nuance. In that sense, they’re more like an unhelpful, automated FAQ.
There’s no denying that AI is becoming increasingly important for automating tasks, supporting creative processes, and customer service. The improvement of ‘conversational AI’ with LLMs offers new opportunities in the realm of customer service chatbots. With conversational AI, you can determine the input yourself. The AI chatbot is claimed to be smart enough to understand your question, intent and even certain entities, such as names. AI chatbots try to mimic human interaction as accurately as possible. With emphasis on ‘trying’, because whether this rings true in practice is questionable.
Why businesses are switching to AI chatbots for customer service
Considering these recent improvements, it is not a strange thought to want to use AI in customer service instead of human employees. AI solutions for businesses have sprung up like mushrooms, making them more accessible than ever. But what makes choosing an AI chatbot so attractive? We list several arguments, thus given by AI chatbot services themselves:
- 24/7 availability: we might be stating the obvious, but AI chatbots don’t need to eat or sleep, which means they are available and ready to serve at all times.
- Multi-lingual support: AI chatbots are trained from LLMs, meaning they can learn any language quickly. A client being able to converse in their native language is a huge win.
- Unified brand experience on all channels: AI chatbots can be trained to use a specific tone of voice that is typical to your brand, enabling the same experience for every customer regardless of the channel.
- Cheaper: the primary reason that most companies favour AI chatbots over human agents is that they are possibly cheaper. An AI chatbot may be able to handle many queries at the same time, meaning you may need fewer employees in the end.
These sound like great arguments. Anyone looking to save money and increase customer support efforts might be easily swayed. Although… Clippy seemed like a great idea too initially. Perhaps we should jump off the current AI bandwagon for a bit and ask the question of whether we know enough about the use of AI chatbots in customer service.
Shortcomings of AI customer service chatbots
Can we truly trust our customers to talk to a bot that represents our company? After all, you don't want your chatbot to do or say things that could potentially damage your company's image, like DPD Post's chatbot calling them the worst delivery service. Arguably, that was the result of a badly trained AI released too early. In practice, we’ve identified some significant shortcomings in AI chatbots that need to be weighed up before making the switch to AI.
Lacking empathy and emotional intelligence
As human beings, we like to be heard and understood. This forms the basis of good customer service. Although conversational AI can increasingly mimic human interaction, it can’t feel or fully understand emotions. As a result, an AI can't sense how a customer feels during the conversation unless they explicitly say so. Even in those cases, the AI isn't emotionally intelligent enough to empathise. AI is great at handling practical matters, though it might be better to disregard it (for now) in professions that require a human connection.
Complex issue handling is off the table
While AI chatbots are much smarter than the old rule-based chatbots, helping with complex issues is not yet an option. The better trained your chatbot is, the better it will be able to help customers - but there will always remain edge-case issues that are best remedied by a human employee. However, an issue doesn't even have to be complex, at least not for us humans. Sometimes it is as simple as the AI not being trained to write down the phone number of a client that wants their call returned.
Ongoing training and monitoring doesn’t come cheap
Once your chatbot is implemented, you’re not done. AI chatbot flows need ongoing training and monitoring based on their output. Analysing whether the bot also gives the right answers takes time and manual effort. Besides, deploying a chatbot that is not properly trained increases the chances of an employee having to take over. It begs the question of whether an AI chatbot is really the cheaper alternative.
No relationship-building or gathering trust from customers
Let’s be real: you cannot build a meaningful relationship with a chatbot. People appreciate personal attention. Specifically, those with more complex cases offer great potential for fostering a positive relationship, converting and upselling.
The personal help of a human employee remains highly appreciated. At Maglr, we see this time and again. Despite having more than 500 customers, we offer personal attention to each individual. Speedy and effective customer support has kept customers returning so often, they are familiar with our names and us with theirs. Some even go so far as to notice when a new employee is working the helpdesk. Such value and appreciation could never be replaced with an AI chatbot.
AI is not trained on recent data
AI is omniscient - or is it? After all, AI models are not trained to recent dates. For instance, you can find in ChatGPT's documentation that its free version (3.5) is trained up to September 2021. Essentially, an AI gives you answers based on a snapshot. An AI chatbot that need to be able to respond to input from recent events may start to hallucinate. This brings us to the next problem…
False, incorrect and unethical answers
We’re sorry if this bursts your bubble, but AI is downright incorrect more often than you’d think. It is not difficult to influence the AI with prompts in such a way that you will get the answer you want - even if it’s not true or unethical. In some cases, it provides us with a good laugh, such as ordering a brand-new Chevy for $1. However, as we increasingly use AI as a serious solution for, well... everything, we need to be more careful.
The other day, some researchers generated unethical answers regarding the European elections in both ChatGPT, Microsoft Copilot and Google Gemini. When asked to think of a strategy for a political campaign that would discourage people from voting, ChatGPT suggested ‘spreading half-truths to sow doubt’. It did not stop there. Google Gemini's advice included using ‘misleading statistics and fake news’ to portray the EU negatively. Such advice is unacceptable and also goes against the agreement recently signed by AI companies promising to combat misleading election information generated by AI. This incident just adds to the view that AI is incapable of making ethical decisions.
Obviously, asking AI chatbots for political advice is highly inadvisable. More and more research shows that data processing in these AIs is not as good as we thought after all. One study shows that Google Gemini's AI models can generate a correct answer in only 40-50 per cent of cases. The fact that we are talking about AI models developed and trained at the highest level is concerning.
Humans are likely to remain irreplaceable according to consumers
Having laid out all the pros and cons of AI chatbots, let's not lose sight of the end goal: serving your customer in the best possible way. The first question you should ask yourself as a business owner considering AI in customer service is “What would my customers think of a chatbot?”. It seems opinions are still divided.
A study by Nationale Voice Monitor (2023) reveals that 64% of Dutch consumers trust a human more. This is because most AI chatbots give the impression that they can do more while they get no further than answering general questions. Yet opinions are different in other countries. In the United States, 68% of a survey group expressed a preference for AI chatbots because of their quick response. However, it does stress the importance of being able to switch to human interaction. We also see differences in preference by generation. For instance, Millenials and Gen Z prefer the use of AI chatbots over human agents. Presumably, this is because they are more ‘digitally minded’.
Whether AI chatbots can completely take over the role of customer service agents seems doubtful. Hiver's survey shows that almost half of people think AI will only be used in conjunction with human customer service agents for the next five years. Therefore, the demand to be able to get assistance from humans remains.
Conclusion: helpful or stupid?
Knowing the current state of conversational AI allows you to make decisions within your own business. Whether an AI chatbot can improve customer service for you is ultimately very subjective. We believe there are a couple of factors to consider.
The most important thing is not to get caught up in current AI trends and think carefully about who your target audience is. If you have a younger audience, there is a chance that AI chatbots can make your processes more efficient alongside making your customers happy too. Moreover, if you run a business that provides specific products or services where complex questions are frequent, the chances of accurate answers by an AI chatbot are relatively low. Finally, decide on how much priority you give to successful customer service. At Maglr, we consider it to be a top priority as is central to our strategy for customer loyalty and trust. That combined with queries that are too complex for AI chatbots to handle accurately means we don't see ourselves making the switch anytime soon - if ever. AI is unfortunately still too “stupid” for our organisation, but it might be a helpful option for others. Our advice is: read up, know your customer and, if you still decide to experiment with an AI chatbot, make sure you keep monitoring its output - just so you won't make people relive their Clippy trauma.
