Businesses get customer feedback through multiple channels – both offline and online feedback. In fact, online feedback systems and platforms that, by design aggregate customer sentiment, are becoming dominant these days with social media platforms. Customers share their thoughts through Facebook likes, Twitter tweets, LinkedIn comments, Pinterest pins, and more.
The real challenge lays in aggregating usable data from these multiple sources, analyzing the data to identify problems, clustering similar problems together, and making them actionable for decision making or problem-solving. This process of converting unstructured voice of customer data into meaningful, coherent datasets to measure customer opinions, product reviews, feedback, or sentiment analysis, and/or subjecting the data to entity modeling in support of fact-based decision making is called Text Analytics or Text mining.
Sentiment analysis, one major division of text analytics, deals with “voice of the customer” materials such as reviews and survey responses, on social media and other online forums. It aims to determine the attitude of the speaker, emotional reactions of the writer, and overall context of the event. The response can be an evaluation, judgment, affective state or emotional communication.
Customer response or opinions may not always be around product reviews. They talk about many instances such as – users are unable to complete their cycle or task due to problems with direct product interactions, software error, or incorrect/ambiguous instructions. For example, users may have a perception that businesses owe low prices only to their loyal customers, and as a result of which they abandon their purchase. A thorough analysis of the customer feedback or behavioral data often reveals the reason for the problem – it could be because of misguided information, an outdated URL, or misinterpreted offers, etc.
Hence, when dealing with the voice of customer data, details on customer journey problems are often detected through direct comments, open text responses to questions like “What is the one thing that would have improved your visit today?” When customers with similar specific problems (responses) are grouped together, product managers and development teams can use their relative volume to prioritize their repair backlog and release calendars.
These kinds of customer problems can be further analysed, drilled-down, classified and ranked by their impact on businesses’ KPIs such as time to purchase, customer satisfaction, etc.
Since the narrative prose written by customers is not structured data, it is resistant to automated analysis, as the combinatorics of human language (the Oxford English Dictionary lists 171,476 unique active words) makes parsing difficult, and manual classification cannot typically cope up with the volumes of text seen by large organizations on a daily basis.
We at Softcrylic have developed an automated algorithm that clusters similar comments from customers, and then evaluates those clusters based on their KPI impact. Similar comments are clustered together based on n-grams sequencing (contiguous sequence of n items from a given sequence of text or speech) that identifies comments that have the highest level of similarity. Comments that use common words, phrases, similar topics, etc. are grouped into clusters and classified based on business logic. Hence, these clusters are often found to be analogous to product categories, customer concerns, product highlights, etc.
For example, the discussion may be around a loyalty benefit that is unavailable to new members, customers unable to edit the cart for certain combination of SKUs, or a newly launched feature that is doing well. Custom dictionaries parse these topics further to identify where the user came across the loyalty misunderstanding. It figures out a list of reasons – the user has not received the welcome email or the customers have misread the information. Each of these topics is then evaluated against the KPI impact and the one with the highest negative impact is identified to make it better with additional UX research, UI design changes, new functional enhancements, etc.
Ongoing analysis additionally supports seasonality, demographics, geographic comparisons, and other segments or data slices as appropriate to the business.
Data extracts from thousands of customer comments that might otherwise lie dormant gathering digital dust, quantify answers to the age-old business question – “Where do I get the most bang for my buck?”