AI chatbots are considerably cheaper than human employees, but they’re not universally well received. How do you maximise their advantage currently – and will technology such as ChatGPT make them better conversationalists in the future?
When you have a million customers and nowhere near that many staff, one-to-one conversations take a lot of bandwidth. And yet, those conversations can be invaluable – whether it’s handling sales queries or resolving customer issues and boosting retention rates.
That’s why, over the years, companies have turned to AI-powered chatbots to engage in one-to-one conversations on their behalf.
The concept of chatbots goes back a long way. As early as the 1950s, mathematician and Enigma codebreaker Alan Turing devised the ‘Turing test’ to determine whether computers could be smart enough to fool people they were human. When the first chatbot, ELIZA, was being developed, the Beatles were still touring.
Nearly 60 years later, and today’s chatbots are way more sophisticated than ELIZA. Recent AI-driven descendants use machine-learning (ML) to enable them to become ‘smarter’ over time and use natural language processing (NLP) to deliver responses that mimic the way people talk.
Proponents will tell you they work exceptionally well, for example by acting as first-line engagement with customers and providing out-of-hours and emergency contact services, which saves big on human hours and associated staff costs.
However, there is a considerable body of opinion claiming they’re not nearly as good as they’re made out to be and, in many instances, they frustrate customers – the opposite of what you are trying to achieve.
Despite the length of time we’ve had these systems, it’s clear that many companies are still getting it wrong. So, how can it be put right?
The good
“There’s evidence that text-driven customer service may be preferable for some customers than speaking with a company rep”
Chatbots, of course, enable you to immediately answer queries, give sales or marketing advice, or handle any issues. There’s no limit to the number of consumers it can interact with, so you don’t lose customers because you can’t get to them in time or they’re frustrated by your lack of availability.
They don’t need sleep, so they’re always on, and there are no theoretical limits to the number of customers they can converse with via speech or text at a time.
There’s even strong evidence that text-driven customer service may actually be preferable for some customers than speaking with a company rep.
According to OpenMarket, 74 per cent of millennials rank text as their most used comms method, 75 per cent avoid phone calls altogether, while 81 per cent get apprehension anxiety if they have to make a call. This speaks very heavily in favour of using text-based chatbots over customer service calls for selected groups – or at least providing the option.
The bad
“More than half report that when the chatbot isn’t able to resolve the issue, they were still unable to connect to a human being – which means their query went unresolved”
In a recent Forrester survey for Cyara, customers gave a thumbs up to certain aspects of chatbot performance, particularly faster response times.
However, this disguises dissatisfaction in other areas. Respondents gave their overall chatbot experiences an average score of 6.4/10, which feels very much in the “could do better” category. Around half reported frustration around their interaction with chatbots, with nearly 40 per cent of these interactions described as negative.
Dig into the reasons and you find three-quarters saying chatbots can’t handle complex queries, almost half saying they’ve received responses that make no sense given the context, while more than half report that when the chatbot isn’t able to resolve the issue, they were still unable to connect to a human being – which means their query went unresolved.
Given that 30 per cent of consumers say that after a bad experience, they are likely to switch to a different brand, abandon any purchase or share negative comments about the brand, this is not something to be taken lightly.
Horses for courses
“Conversational dialogue, where the topic is relatively open ended, is trickier, hence the development work and excitement around new AIs”
Chatbots are good at some kinds of interaction and bad at others. To assess how well AI does at mimicking humans, and resolving issues as humans would, it may be useful to differentiate between “goal-oriented dialogue” and “conversational dialogue”.
The former is when you are talking to the AI to achieve a specific goal – making a restaurant recommendation or getting a digital assistant to play a piece of music, for example. Although not without the odd hiccup and unasked-for suggestion, AI is pretty good at these things.
Conversational dialogue, where the topic is relatively open ended, is trickier, hence the development work and excitement around new AIs. When queries are straightforward and on-script, a good chatbot will usually rise to the occasion. When the use case gets complicated, the chatbot can soon get lost or turn into an inflexible jobsworth.
Human beings can say the same words in a different tone or in a different context and mean entirely different things. Even other humans are bad at detecting written emotion or irony. However, we’re typically pretty good at spotting it in conversation. Chatbots, on the other hand, often suck at it, which can lead to huge frustrations.
Working with customers
“For chatbots to work effectively, we need to play to their strengths”
For chatbots to deliver savings and convenience to brands, while providing a great service to their customers (and not driving them away), the key things are to let the customer know immediately they are talking to a bot so they can do whatever they need to do more quickly. However, they also need to know that they can switch to talking to a person at any time (or at least during local trading hours, depending on your business).
For chatbots to work effectively, we need to play to their strengths, which (at the moment) is far from every job. If you’re looking to AI as a replacement for human support rather than a supplement, you’re likely to lose customers.
More sophisticated systems, such as ChatGPT, hold the promise of AIs being able to handle more of the workload, and we are likely to experience for ourselves how good they actually are in the immediate future.
However, few people expect AIs to replace human reps any time soon. It is more likely they will take on a greater workload, which ideally should free your customer support staff to handle the more complex (and potentially interesting) tasks, or those requiring a degree of emotional intelligence that AI just hasn’t got to grips with yet.
Want more insights from ITG CEO & founder, Simon Ward? Discover the full Simon Says series here, and keep an eye out for the next entry coming soon! And remember, you can contact us about our integrated, end-to-end creative services and Storyteq marketing technology any time – just drop us a line at hello@inspiredthinking.group or fill in the form below.
Fill in the form and we’ll get back to you.