How machine learning could improve CX intelligence
With growing volumes and varieties of available data, businesses are increasingly keen to extract valuable information so they can use that knowledge to make predictions, optimise decisions and develop policies.
This is particularly the case in the customer service industry, including call or contact centres, where organisations are seeking to improve the customer experience by investigating what and how their behavior influences customer evaluations on their business.
The technology behind these data extraction innovations is known as machine learning, which I’ve been researching for the past 10 years. Machine learning has been used to develop driverless cars and effective web search among many other innovations, and is the science of getting computers to act without being explicitly programmed. It is believed by many researchers to be the best way to make progress towards human-level artificial intelligence.
Extracting valuable information from data often resorts to machine learning techniques. Machine learning algorithms can automate the data analysis process by applying complex mathematical models to discover hidden insights from the data, and are core elements in big data analytics.
Customer experience is an important measurement of service provided by the organisation. This is said to be a transformation from customer-organistion interactions to customer personal feelings on the interactions. However, it is not easy to establish the rationale for such links given the limited numbers of interactions that often don’t come with a label of customers’ evaluation. More seriously, customer evaluation is often context-sensitive and the interaction data doesn’t record all relevant factors or features.
With a wealth of data available in call or contact centres, SMS message and short-chats in existing social networking platforms, we aim to find a set of informative features that fully capture the quality of interactions, and then develop predictive models for customer experience improvement.
We expect that the models could explicitly articulate associations between the identified features and the customer evaluation on the interactions – organisations could find key features that impact customer experience, and react accordingly.
And this expertise is in high demand from businesses. I am currently involved in a Knowledge Transfer Partnership (KTP) funded by Innovate UK with a company that specialises in recording calls from businesses’ contact centres, providing clients with the ability to retrieve, score and analyse performance, to find ways to improve the customer experience of the service they deliver.
By applying these principles to the customer experience market, we are looking to transform call-recording functionality via software that will help businesses to record and score every element of a call – from the agent’s tone of voice to time waiting on hold while a call is re-routed. That data will then be available for automatic retrieval and analysis.
The project will first fuse data from different types of communication channels that record interactions between agents and customers in different time intervals. Then we will identify a set of features that are most relevant to customer evaluations on the interactions. Subsequently we can develop a predictive model that relates the features to the customer evaluations. The creation of this model is a backbone of this project and will be incrementally refined upon new data and additional business knowledge.
Our work will implement a customer experience evaluation component and ultimately have it embedded in a commercial product whose customers include many major banks and high-technology traders. With more data retrieved from various channels, the project will incrementally improve the product on precisely scoring the customer experience therefore resulting in delivering high quality of service to the clients.
More and more of this type of work is expected in call or contact centres, with companies able to achieve new insights and learnings that would formerly have been invisible to the human eye. We believe that machine learning techniques will play more and more important roles in analysing customer interaction data that have been accumulated in call centres over years. They will not only facilitate customer experience improvement, but gain more insights about customers’ behaviour in a wide range of contexts. Meanwhile, the techniques may provide guidelines on how the interaction data should be recorded, maintained and fused, which is often limited by physical, computational resources.
We expect that machine learning techniques will lead to a new generation of service software-agents who will interact with customers in a smart way. More importantly, these techniques may help personalise the agents who will show human-like social behaviors in the interactions. With such, there will be un-manned call centres that seamlessly links customers with service, and immediately bring economical, social benefits to businesses adopting them.
Yifeng Zeng is a Reader in School of Computing at Teesside University.
Dr. Zeng is a Reader in the School of Computing at Teesside University. He received his PhD in 2006 from the National University of Singapore, Singapore. Before he moved to Teesside University, Dr. Zeng was an Assistant Professor and an Associate Professor during 2006 –2012 in Aalborg University, Denmark.
His current research...