skip to content

Cambridge Service Alliance

At the forefront of service transformation in the digital era

Successful customer service feels personal. As customers, we need to have our problems dealt with efficiently. But we also want our emotional response to those problems to be acknowledged, with empathy.

But is this possible in a world in which customer services are becoming increasingly automated, and we find ourselves conversing with chatbots? Our research considers how AI may make it possible for digital agents to respond appropriately to customers’ needs and emotions and, by showing ‘empathy’, deliver excellent customer service.

The rise of personalisation

We know that people want a truly personalised service. Some recent (2021) market research suggested that 79% of customers think personalised customer service is more important than personalised marketing or product recommendations. Yet it is the latter which has attracted most attention – and industry plaudits - with the likes of Amazon and Netflix using personal recommendations as a hugely successful means of growing their businesses.

Now, however, as we take the convenience of online services increasingly for granted, some of their drawbacks are becoming more apparent. When a transaction does not go smoothly, having to interact with a chatbot can significantly add to a customer’s frustration. Where a human customer service provider can detect that frustration and help to alleviate it, a chatbot has not, to date, been equipped to do so.

Our research considers the potential role of AI technologies in overcoming these challenges. Can we, by using everything we know about both a particular transaction and the customer engaged in that transaction, enable the chatbot to interact meaningfully with the individual, rather than supply generic responses to questions?

To understand how we might go about this, we explored the available literature in psychology, customer experience, customer service and computer science. We divided the task into two parts: how can we acquire the knowledge we need about the customer and then how can we design the communication.

First, know your customer

Customers’ emotions arise from three key factors relating to a particular event: how that event played out (including how an issue was resolved and what their previous experience with that company had been like), how the agent communicated with the customer and how effective was the interface? Each of these three factors can evoke a wide variety of emotional responses, all of which need to be managed.

Understanding emotions

While emotions are short-lived and triggered by an event, moods are not event-specific and can last for longer periods of time. For the purposes of improving customer service, responding to emotions has been shown to strengthen the relationship between the customer and the customer service provider. In order to react appropriately, the customer service agent must be able to distinguish between different types of emotion.

Like with like

However, identifying the emotion may not be enough information for the agent to respond appropriately. A customer’s emotional response to a problem will be determined by their personality type. But how can we find that out? Research suggests that it may be possible by analysing their public domain activities such as social media posts.

Another well-supported theory is that people tend to be attracted to people like them. For our digital agents to be well-received, therefore, they need to respond in a way that matches the customers’ personality traits.

Previous experience

How a customer has got on in the past is going to colour their response to any issue. Knowing the customer’stransaction history and their past behaviour will also help to create an appropriately personalised response.

Second, strike the right note Humans are social beings. When designing interactions we need to understand the power of the network, and how much customers are influenced by other people’s interactions with the organisation. Customer reviews are important in this regard, not just in relation to how good or bad a product or service is rated but they also offer social cues as to how the customer experienced the service they received. If others are an important part of the customer journey, how can AI agents step into those shoes?

To do this they need to go beyond accomplishing tasks to engage them socially by displaying, for example, a proactive approach, a positive attitude and courtesy.

Being able to show empathy is a key part of a successful customer experience. It can establish a connection between the agent and the customer. We looked at the theory of emotional contagion to understand how empathetic responses work, through a flow of emotions “with the receiver ‘catching’ the emotions the emotions the sender displays.” Using the right language is critical in conveying the right emotions, as are the right voice and tone and facial expressions and gestures, in the case of an avatar.

Most digital agents either use natural responses or rely on scripts. Natural responses allow the agent to adjust their responses to the situation they encounter. However, they are also more likely to say something inappropriate – and they are, of course, more difficult to generate.

How can all this be achieved?

We looked at different types of AI applications and their potential for supporting personalised customer service responses.

Conversational analytics

This approach is mostly powered by natural language understanding, which allows machines to understand and interpret data from human language. Researchers are already using advances in this field to detect customers’ emotions through call centre conversations or email exchanges. However, most are still fairly broad brush, identifying positive or negative emotions but not able to undertake a more fine-grained analysis. Nonetheless, we are starting to see products on the market offering solutions which can detect customer emotions and personality.

Conversational analytics can be applied to previous customer interactions across different channels (such as phone, SMS, chatbot and social media) to build up a more detailed picture of the individual and their progress along this particular customer journey as well as previous one. Social media and publically available information about the customer can help the agent understand where the customer fits in terms of social groups and hence predict likely motivations, preferences and behaviour.  However, while social media is often screened for customer feedback, to date it has been less widely used to gather this kind of information and apply it to customer service.

Conversational coaching

Conversational coaching builds on conversational analytics by combining natural language understanding with deep learning algorithms to provide the agent with real-time suggestions of how to improve their interaction with the customer. These kinds of systems are already being used to, for example, to provide personality-based services or suggest socially-appropriate tourist attractions. But they are not yet used to empower a digital service agent to have personalised conversations with customers.


Advanced chatbots need natural language understanding to understand the customer and natural language generation to respond. Recent advances in this field have delivered outstanding performance in answering questions, summarising text and analysing sentiment. These capabilities could be used in chatbot applications, giving them the ability to deliver great service. However, chatbots suffer from one significant limitation: their lack of a ‘physical’ presence prevents them from displaying emotion. 

Currently, chatbots are successfully using analytics to detect a customer’s personality and emotion. The next step is to enable them to respond empathetically, using the right strategies for individual customers.

Virtual avatars

Giving a digital agent a human form enables them to display emotions using facial expressions and gestures. This technology is still in its relative infancy with significant challenges still to be overcome, not least allowing for wide variations in different interpretations of how emotions are displayed. But there are some examples such as Microsoft’s virtual agent Xiaoice which was developed to help people with loneliness and provide emotional bonds through a warm-hearted conversation.

Insights for managers

What does this research tell us? Until now, firms have used digital agents as a way of expanding their customer service capabilities while controlling costs. However, it is becoming increasingly clear that advances in AI promise the possibility of personalised and empathetic – but automated – customer service.

There is still some way to go before that prospect becomes a reality but if it does, the prize will be significant: increased customer engagement, satisfaction and, ultimately, loyalty.

Jan Bluemel, PhD student, Cambridge Service Alliance
Professor Mohamed Zaki, Deputy Director, Cambridge Service Alliance.
Dr Thomas Bohné is the founder and head of the Cyber-Human Lab at the University of Cambridge’s Department of Engineering

Picture Credit: Tero Vesalainen

Cambridge Service Alliance

Welcome to the Cambridge Service Alliance…

  • A unique global alliance between the University of Cambridge and some of the world’s leading businesses.

  • Help organisations to address the challenges they will face in the next three to five years, through rigorous research, practical tools, insights and education programmes.

  • Learn how other innovative organisations are developing new services through our events

  • Since its inception in 2010 industrial partners have included CEMEX, GEA, IBM, Pearson, Zoetis, HCLTech, Bouygues UK among others.