analysis of complaints analysis of reservation changes

InteliWISE introduces machine-based analytics

Decision makers have always been interested in their customers questions and this issue is the basis for increasingly intelligent analysis. Instead of guessing what customers expect, companies can now use InteliWISE analytics. This solution will tell you exactly what critical problems need to be addressed and what areas should be automated e.g. handed over to a Chatbot.

The growing popularity of chats-based contact requires sharper, more intelligent analytical tools. 

Chats are growing in importance in marketing communications and customer service they often respond to 20-30% of all queries. Calls are handled by consultants via Live Chat, Messenger, self-service forms or 100% automatically via Chatbots, i.e. Virtual Assistants.  

The challenge for marketers or customer service experts is to analyze them. It’s easy to check how many chats consultants support every hour or day but the topics of conversation or the vocabulary used in hundreds or thousands of chats or emails remain unmeasured.

Artificial Intelligence supports reporting

And here is where Machine Learning algorithms come to the rescue. InteliWISE offers companies our Topic Modeling Analysis solution – a new, advanced analytics tool for conversations with clients based on quantitative methods and NLP (natural language processing). This new tool extends the methods used by Machine Learning and NLP with a tool from the Topic  Model category. It is a solution that helps to search for abstract, set topics found in large collections of often unstructured data. Its effect is often the discovery of seemingly invisible but significant semantic structures in the text. All you have to do is provide historical customer queries (chats, emails, Chatbot conversations), and the analysis tool will generate an accurate picture of what and how often customers ask about specific terms or topics, including:  

  • a list of specific questions and conversation topics that customers present, along with their relative importance
  • exact percentages of chats that fall into selected topics (the remaining sessions constitute chats that are almost impossible to categorize) 
  •  chat dynamics , tracked on the basis of comparisons.

Topics, problems and their relative weights are presented in the form of pictographic charts. 

Sample analysis results presented can be numerical metrics, most frequently chosen topics below: analysis of reservation changes

analysis of reservation changes

Sample analysis result related to a narrow, less numerically specialized group of topics below: analysis of complaints

analysis of complaints

This analysis allows not only to understand the basis of 70-80% of problems, but also to understand these hidden relationships meanings. In business terms, this can translate into a reduction in service costs and significantly better Customer Experience ratings.

How can a company use the new solution for analyzing the operation of Chatbot and Live Chat?

Traditionally, after talking to the client, consultants decide (qualify) the nature of the conversation and decide on the next steps. This new method allows you to free yourself from consultants subjective view.

How we do it

1. The Ordering Party provides a CSV file or another with historical conversations with clients (via Live Chat, Chatbot, Facebook or application). 

2. We analyze chat sessions from a given period. For this we use machine learning tools,  thereby enabling effective analysis of large volumes of queries (libraries of historical conversations). Each of the chat entries is processed for normalization for further processing by: 

  • removing popular words in any language that do not carry meaningful  content (e.g. “up”, “in”)
  • simplifying words  and removing punctuation,
  • removing polite phrases and other placeholders , which often occur during a Live Chat session,
  • finding word combinations that together form items or action names such as “carry-on luggage.”

3. Live Chat sessions prepared in this way are processed by a model based on latent Dirichlet Allocation (LDA), which allows searching for a set of topics in large batches of unstructured text. 

4. The result of the algorithm is to find a set (measured in percent) of popular topics, each of which is characterized by a set of keywords in its vicinity (i.e. in the same context). This is a collection of chat sessions that have been qualified to selected topics. The remaining sessions concern cases so rare that their qualification is impossible. On the charts, the size of the circle represents the popularity of the topic. The space between the wheels indicates the degree of relatedness between the given issues.

Understandable analysis results allow easier intelligent automation

Companies serving hundreds of thousands of customers, such as significant online stores, banks or insurers, have a significant problem with understanding how automation of query handling can work. But thanks to InteliWISE, companies can base their decisions on objective AI methods that are clear and cost-effective.