Using AI to improve customer satisfaction and reduce churn
Businesses know quite well that it costs less to retain a customer than to acquire a new one. In this current environment there is no doubt that customer retention should be high on every CEOs list. Given the pace of technology in the Artificial Intelligence (AI) arena there is no better time and no doubt that AI, and more particular Machine Learning (ML), can help in this regard.
First a quick note on machine learning current evolution. Gone are the days where machine learning was accessible to the select few large enterprises with large data centres full of computing power and rooms of data scientists. In today’s world of elastic cloud computing, enterprises of all sizes have access to computing resources from customizable field programmable gate arrays (FPGAs) for large custom computations to high memory instances to compute intensive GPUs. In addition, many cloud platforms have embraced the new ways of consuming and transforming data and the open source world full of libraries and projects, making them accessible to everyone to easily build custom machine learning models. Finally, operationalising AI by deploying and even retraining models when new data comes along can be automated using cloud-based CI/CD pipelines allowing for smaller more business focused data scientists.
Customers churn for many reasons, some for lack of service and some for financial reasons but the outcome is the same, your competition is now the proud new owner of your hard-earned customer. So how do we keep those customers from switching? That’s the easy part, by listening to them – through data.
Customer experience can be made up of many touch points including purchasing the service, consuming the service, paying the fee, and calling the call centre. Each of these encounters leaves a data footprint that can be used to create a “picture” of the customer and the more the customer interacts the better the picture becomes. But the picture is never static as it evolves over time and it is that evolution that can tell us how happy the customer really is at a specific point in time.
However, with thousands of customers and millions of individual customer transactions it is impossible for account managers and customer service agents to truly get and understanding of the customer. Some of the tell-tale signs include no longer consuming the service, no longer paying the fees, and continuously calling the call centre over a short period of time.
To keep track of customer experience is a mammoth task and co-ordinating this data from separate sources is one of the biggest challenges but with the advent of data streaming technologies, data lakes and the ability to automate data process using Extract Transform and Load (ETL) pipelines the task is not insurmountable.
That said, some customer data does not exist as yet and needs to be developed. This could be by running analytics on the data for example calculating the mean time between calls at the call centre or the number of payments missed. But some data is a little trickier to develop, for example, how did the customer feel on the last call. Yes, there is a chance that the customer rated the service at the end of the call but many of us never take the time to do that. A much more effective way is to evaluate every call using AI, using a process called sentiment analysis in a field called Voice of the Customer (VOC). Using AI, even in real time, the call is transcribed using new branch of machine learning called deep learning to recognize speech and convert it to data in the form of text. Following that, another arm of machine learning called natural language processing (NLP) can be used to get insights into the call like the sentiment of the customer or even identify key words or phrases. Now the enterprise has a real time view of how the customer feels on every call and whether they are a satisfied customer or not.
But AI can go further, we can even have a much better understanding of the customer by combining the data we have and data we have created, and with that we can take a final step and build a machine learning model to predict whether a customer is likely to remain or churn. For this type of business problem, a machine learning classification algorithm is ideal and can be built to find patterns using historical data of customers that we know to have actually churned in the past. While building the model could take as long as a few weeks to train and tune the performance of the model, once the model is deployed the predictions can be obtained in near real time.
Unlike the backward-looking time delayed customer satisfaction surveys, retention teams or a customer service representative can take real time pro-active actions, based on the predictive analysis, to retain the customer and secure the ongoing revenue for the company.
