How will the next generation of AI bots transform CX?

Who wants to talk to nobody? Are customers ready for it?

If, like millions of others, you’ve tried using the new AI bot ChatGPT, you will have to agree it’s an impressive piece of technology.

Unlike most, if not all, other chatbots it just gets what you tell it and responds in an incredibly human-like way. It might sometimes get its facts wrong, but it rarely misunderstands your meaning or intention.

ChatGPT is also ridiculously intuitive to use. There is no traditional user interface to learn – you just start talking (well, typing).

The intriguing question for those of us that work in the CX and contact centre space is how will technologies like ChatGPT be deployed to help brands connect more deeply and more easily with their customers?

Now that this thing is out of the box, and millions of people have used it, will they be less willing to put up with today’s unintelligent bots that are designed to handle a limited number of customer queries?

Will they want to talk to, well, nobody, if it can help them solve their problem more easily? Or will they still prefer to talk to a fellow human?

Seeing the future

ChatGPT is a generative AI language model. It works by analysing an input (i.e. what the user types or says) and produces a response by trying to predict what the next word should be. That might sound quite simple, almost limiting, but as you can testify if you’ve tried to use it, ChatGPT can write about almost any subject, in any language, and in almost any style or tone of voice you want.

We asked ChatGPT if we could use it to build a chatbot for a contact centre. It not only answered the question, it gave us several use cases and then broke down – step by step – exactly how we would go about doing it.

We asked ChatGPT if we could build a ChatGPT chatbot for a contact centre. Here is its unedited response:

Yes, it is possible to build a chatbot for a call center using ChatGPT. However, building a chatbot using ChatGPT requires technical expertise in natural language processing, machine learning, and software engineering. You would also need to have a large dataset of high-quality labeled data to train your ChatGPT model.

Building a chatbot for a call center involves understanding the specific needs of the business and its customers. The chatbot should be able to answer frequently asked questions, provide information on products and services, and handle customer complaints and inquiries. The chatbot should also be able to escalate complex issues to a human agent if necessary.

To build a ChatGPT-based chatbot for a call center, you would need to follow these steps:

Overall, building a ChatGPT-based chatbot for a call center can be a complex and time-consuming process. It requires technical expertise and a significant amount of data to train the model. However, if done correctly, a ChatGPT-based chatbot can improve customer service and reduce the workload on human agents.

Note that one thing it doesn’t do is preserve the Australian English spelling we used to ask the question – it reverts to US spelling automatically. Other than that, it’s a coherent, clear, detailed, and very honest response.

As is always the case with transformative technologies, we can only see so far into the future. Generative AI models are still relatively new and improving at a rapid rate. This means there will be lots of use cases we just can’t anticipate until someone invents them.

So we are keeping our predictions to the next five years, as if we look much further beyond that we may as well be writing science fiction.

So, here are our top 5 predictions.

1. Chatbots become genuinely useful

The majority of chatbots that have been deployed by contact centres to date aren’t actually intelligent. Sure, they are built on NLP (Natural Language Processing) technology and include features like text-to-speech and speech-to-text, however, their intent matching models are rudimentary compared to something like ChatGPT. The datasets they are trained on are infinitely smaller, which means they are only able to respond to queries about a very limited range of topics. Most of them, however, don’t include much ‘intelligence’ at all, and are just pre-programmed to give canned responses to the handful of queries they have been trained to recognise.

ChatGPT is different. It hasn’t been programmed to recognise a certain set of intents. It doesn’t work in a controlled environment like the customer service department of a business that only ever deals with a finite set of queries (the odd outlier excepted). Anyone can ask it any question they like and ask it to perform an almost unlimited range of complex tasks. It can even write perfect working computer code, in any computer language, in response to nothing more than a written description of what the code is supposed to do.

Over the next few years (if not months), we will see new chatbot products based on ChatGPT and other generative AI models released by both the major cloud contact centre vendors and niche players who will be responding to specific needs in different sectors. These bots might not have the full range of knowledge that ChatGPT has – they wouldn’t need it – but they would have all of its incredible language parsing, intent matching, and response generating capabilities. Instead of being trained on the whole of Wikipedia they would be trained to know everything it is possible to know about the operations and customer interactions of the organisation deploying them.

2. Automate more queries and follow-up actions

In theory at least, a chatbot built on ChatGPT technology wouldn’t need to be pre-programmed to recognise every possible customer query and taught how to respond to each one. It would just understand and respond. If it was hooked up to some RPA (robotic process automation) bots it could even carry out the actions needed to solve the customer’s issue with no human intervention. For example, it could apply a credit to a customer account, process an order, take a payment, update records, book an appointment, or anything else a human advisor can do during a customer interaction, or in wrap afterwards.

With its quite stunning writing capabilities it would also be able to reply to customer emails in seconds with a totally original, personalised, and accurate response, while triggering any necessary follow-up activities, and updating any relevant records and notes in the appropriate databases. With the almost realistic-sounding text-to-speech capabilities that chatbots already have, a ChatGPT-based bot will be able to answer voice queries just as easily as webchats and emails. If it does get the wrong end of the stick in any interaction, unlike existing chatbots it doesn’t just get stuck or have to hand over to a human agent. It can ask clarifying questions and update its understanding, just like a human advisor would.

3. Help human agents do their job better

In the immediate future, we are more likely to see generative AI bots in the contact centre as tools to support human advisors. There are several areas ripe for immediate application to help advisors respond to customers more quickly, efficiently, and accurately.

Bots based on generative AI technology can help advisors by providing them with information at appropriate times during customer interactions. Essentially the bot can monitor the call or chat with the customer and suggest responses or next best actions. It could assist with troubleshooting, as it can analyse information and suggest responses, making it a valuable diagnostic tool. Bots could even write draft replies that advisors can edit.

AI can also provide coaching and simulation tools to assist advisors in improving their performance. For example, they could role play different voice and chat scenarios with a bot, which would make up realistic questions and responses based on its deep understanding of hundreds of thousands of past customer interactions.

The language analytics capabilities of something like ChatGPT could also be used to monitor interactions in real-time for quality assurance and compliance, sentiment, and customer feedback. Supervisors could be sent automatic alerts to help out an advisor who was really struggling with an issue. On top of the current knowledge transfer processes, this could be a huge advantage in a remote working environment, where advisors don’t have the back up of a floor-walking supervisor to turn to.

Another application of AI is to help make responses to customers more personalised. By understanding the context around a customer interaction, and analysing historical data, a bot can provide tailored recommendations and solutions to advisors, improving the customer experience. It could also identify upsell and cross-sell opportunities, leading to increased revenue.

4. There will be more roadblocks than we can foresee

It’s easy to get carried away with the possibilities AI can bring, but one thing we know from the rollouts of pretty much all technologies, is that there will be snags. Progress may not always be as fast and as linear as anticipated.

The very fact that a generative AI natural language model like ChatGPT exists, doesn’t mean you can deploy it in your contact centre overnight. ChatGPT is only capable of holding a human-like conversation on almost any topic and writing at least as well as the average University graduate because it takes an enormous amount of computational power, and it has been trained on a lot (A LOT) of data.

To deploy your own version of ChatGPT you would need to train it on your company data, systems and processes. Unfortunately, that requires quite a bit more work than giving it a few training manuals to read. To deploy something as powerful as ChatGPT in your contact centre you need to train it on an extensive, structured data set including recordings of as many past customer interactions as you can find. Those, however, have to be transcribed, checked, labelled, and tokenised (broken up into small semantic chunks to feed to the AI).

The point is that training a ChatGPT-based bot to be sufficiently reliable and accurate to interact with your customers on its own, about almost any topic they might bring up, is a significant undertaking. Few organisations, if any, are in a shape to even begin such a project. It’s why the first generations of bots based on generative AI language models will still be quite limited in their scope and application – not because they lack capabilities (as today’s bots do) but because training them is really hard and is best taken step by step.

5. Customers still want the human touch

As we just saw, it will take a long time for any contact centre to be in shape to deploy a ‘universal’ chatbot for customer service.

We would go further than that though. For now, at least, people still want to talk to another human being about complex or emotional issues, not a robot. The findings from the latest Contact Babel survey into CX were astonishingly clear: 72% of customers of all age groups prefer to talk to a human being rather than use self-service. Even in the youngest digital-native cohorts, which we would presume to be the most tech-savvy, only 21% preferred self-service.

Data taken from Contact Babel

With more capable bots, businesses and contact centres will be able to automate even more complex customer queries than they are able to today, however this won’t mean all customer contact will get automated. It’s just not what customers want. At least, not yet.

Final thoughts

The generative AI language processing technology that powers ChatGPT and the upcoming generation of chatbots and virtual assistants, is a game-changing technology. There is no doubt about that. When we eventually look back, it will be seen as the technology that heralded this new world and awoke us to its possibilities. But there is a long way still to go from this point.

For contact centre and CX professionals, this next generation of bots are going to be used primarily to support human advisors. They will improve response times, accuracy, and efficiency, while also enhancing the customer experience and reducing the cost to serve. Many more common tasks will be automated, and by making customer service more personalised, brands will be able to differentiate themselves.

As the technology continues to evolve, there is the potential for it to take over more of the customer experience but there are two major roadblocks to that happening on a large scale. The first one is that deploying a genuine AI bot is not a trivial undertaking as the AI needs to be trained on vast amounts of structured and labelled data. That takes, currently, a lot of time, money, and effort, although we anticipate there will be a huge growth in technology to automate that process. Secondly, and perhaps even more importantly, customers prefer to talk to humans. Until that changes, there will always be a resistance to talking to nobody.

Let’s leave the final word to the star of the show, ChatGPT itself. We asked it to tell us, in 200 words or less, how it will be used in contact centres to improve the customer experience. Here is its unedited response:

Generative AI technology can be used in contact centers in a variety of ways to improve the customer experience. Here are a few examples:

So, there you have it.

TSA Group is, as you would expect, leading the way in exploring how AI will impact the operation of the contact centre and the delivery of CX.

Our proprietary cloud contact centre platform already integrates multiple automation technologies, and with our partners, we are adding the power of generative AI language processing. Talk to us about the possibilities.

TSA are Australia’s market leading specialists in CX Consultancy and Contact Centre Services. We are passionate about revolutionising the way brands connect with Australians. How? By combining our local expertise with the most sophisticated customer experience technology on earth, and delivering with an expert team of customer service consultants who know exactly how to help brands care for their customers.