Article

Using aI in customer communications

Authors

If you want to leverage the value of modern artificial intelligence, customer communication is one of the easiest areas to start with.

Customer communication exhibits several traits that makes it very well suited for using artificial intelligence:

  • The offline idea that managers or co-workers through co-listening can be on top of what is truly happening is an illusion. A frontline manager with 15 employees can most likely spend no more than an hour a week co-listening, and most managers spend significantly less listening to employees projecting their very best behaviour.
  • Customer communication is incredibly complex. The core data is communication; whether through text or speech, it is language. Around this might be metadata like time, duration, request type, customer ID and so on, but the real gold lies in what is said or written, which is traditionally very hard for businesses to analyse.
  • The volume of customer interaction is huge. Consumer-facing companies have thousands, if not millions, of touchpoints with customers. The volume of data makes it ideal for the power of AI to discover complex patterns and realise benefits once a useful algorithm has been developed.
  • It is valuable. Companies that consistently deliver quality customer service in every interaction outperform their competitors. Communication is a core of the customer experience, and responsive, trained and coached service agents play a vital role in this.

What can you do today?

To approach a customer communication project using artificial intelligence, you need an explorative mindset. You should not just use artificial intelligence because you can, but rather because you want to improve your customer experience, build skills for your employees, reduce cost etc.

You can do a number of things today if you want to leverage your customer communication with AI. Below, we have listed some simple modelling examples that we used working with the intersection between customer communication and artificial intelligence.

Topic modelling

Topic modelling takes a large set of conversations, and an algorithm learns a structure. The structure then sections the conversations into clusters in different ways and allows you to identify topics across conversations.

Above, you can see an example of how keywords across conversations might cluster, and each might “load” on a given topic, allowing you to identify which topics are present in which conversations. From here, you can identify which topics take up the most time, the most tickets in the system or lead to the biggest frustration, allowing us to focus training and better establish a knowledge database etc.

Sentiment analysis

Speaking of frustration, another powerful tool is sentiment analysis. This tool allows you to identify emotions in conversations. Fundamentally, it goes from happy to sad/angry across neutral, but more sophisticated solutions also exist to determine matters of intensity and valence.

Topic modelling and sentiment analysis are fundamental techniques, and both enrich reporting about conversations. Using these techniques together with traditional information on conversation length, time of day and seasonality, you are able to get a real-time deep-dive into what your customers think and desire.

Custom solutions

Finally, there is also the custom solutions. As an example, we worked with a client who was interested in the degree of enquiry from their representatives, and how curiosity and enquiry from their representatives affected conversation flow and satisfaction. Together, we developed an algorithm to detect the degree of enquiry from the representatives and added this value to each conversation, enabling deeper analysis across all dimensions.

Similarly, a common analysis you can make is looking at adherence to a particular operating model or standard for conversations. Well-known touchpoints that all conversations should cover or questions which must be asked. These are relatively easy to identify using artificial intelligence and then scale up to cover thousands of conversations in a matter of seconds or minutes to give you a complete picture across departments, ticket types, time of day and so on.

Then you need to scale fast!

We firmly believe in an agile approach to using AI in customer communication.

You can use the above examples as a proof of value (not just the usual proof of concept) and from there you need to scale your use of AI building on the insights you have established. Do you use it for training of employees? Do you help them access the knowledge database? Do you build performance metrics reflecting more than just metadata? Etc.

Once you have started, you will discover endless opportunities.

If any of this has peaked your interest, do not hesitate to reach out to learn more.