In my previous blog post, I looked at complaints data within financial services and how it can be difficult to elicit a clear picture of what is really happening from top-line information alone. Text can be hard to interpret compared to numbers. But one way of extracting more clarity and granularity is through the use of text analytics. In my second post, I will look at this in more detail and share some real-life examples of how it can help organisations.
Data scientist vs linguist
As I highlighted in my previous post, the “people component” is vital. An analytics or data science team complete most data analysis. Most data scientists come from a mathematical, statistical or general analytical background. As I explain below, the analysis of text, at least the way SAS does it, combines machine learning and linguistic rules.
Whilst data science/analytics teams have analytics experience, it is less common to also have an understanding of linguistics (the study of language). The skill set for strong text analysis is, in my opinion, simply not there. You can usually get a person who builds a very good machine learning model or someone who has an exceptional understanding of languages and linguistics – but seldom both.
How to address this challenge
Certain analytical solutions give you the ability to overcome some of the hurdles mentioned above. They can ingest text data of any format and analyse the "messy text" to conduct root cause analysis to identify, for example, what customers are complaining about. It is even better if you can deploy the solution easily and at scale, both in the cloud or on-premises.
SAS Visual Text Analytics uses a hybrid approach to analyse unstructured data, which encompasses both machine learning (the analytics bit) and linguistic rules (the language bit). Before that stage, it applies natural language processing techniques to get structured information from the unstructured data. It deals with misspellings, applies synonyms (groups words with similar or the same meanings) and takes account of abbreviations. It also uses part-of-speech tagging for text, breaking down a sentence such as "Vanessa is using SAS to analyse data" into its components: nouns, pronouns, verbs, etc.
Once the solution organises the data with NLP, it uses machine learning to create topics, clustering common themes in the data. It will look at the relationship between terms and, in this example, particularly customer behaviour. The solution can extract concepts or entities and key elements of the data, such as addresses, postal codes and dates. It can create categories (i.e., rule-based grouping). Then you can apply sentiment analysis to look at the tone of the text, ascertaining whether the comment is positive, negative or neutral.
Here are some successful case studies within the banking and insurance industry applying analytics to unstructured text data.
Case Study 1: Analysis of complaints in insurance
A major UK insurance company was losing thousands of policies every year due to customer complaints. It was interested in finding out which type of complaint drove higher attrition rates and which customers were likely to leave based on the wording of their complaints. It used text analytics to categorise the complaints. The analysis also allowed it to see the severity of the complaints. This ranged from trivial issues relating to the website to more serious ones, such as broken promises from call centres. It subsequently built predictive models to assess those customers who were more likely to churn based on the complaint they made. Then it implemented a retention programme using the results.
Case Study 2: Analysis of webchat within a building society
Recent growth in members had put pressure on Nationwide Building Society’s contact centre. Receiving over 800,00 calls from members every month, the Society was keen to look at reducing call volumes. It was aware that members often contacted them via other channels. These included browser-based messaging, email and – if their queries were not resolved adequately – they resorted to calling.
Using text analytics, the organisation identified the root cause of each enquiry, whether the query was resolved and how many emails were exchanged. Perhaps unsurprisingly it verified that the sentiment became more negative the more messages that were exchanged. The value from text analytics was immediate. Nationwide concluded that more than half of the messages sent could have been avoided entirely. One easy way was simply to make the information more easily accessible from the website. As a result, it was able to put in place some immediate quick wins.
Do you know what your customers are saying about you? NLP and text analytics give you the ability to learn exactly that. Click To Tweet