Unlocking Business Insights with Text Analytics
Text Analytics in Business
Today, customers interact with companies through multiple channels, making personalized experiences and customer satisfaction more important than ever. Through survey comments, complaint reports, pre- and post-service technical reports, customer service calls, social media, and more, customers communicate with companies to express their perception of product and service quality, submit claims, share opinions about the brand, recommend products, and so on. Because of this, it is estimated that 80% of relevant business information originates in an unstructured form, mainly as text, and that this unstructured content grows at a much faster rate than structured data.
A text is more than just a collection of words—it tells a story. Text Analytics is an area of data science that enables the automatic extraction of useful information from a set of documents or texts. In general terms, text analytics seeks ideas, concepts, and relationships that are “hidden” among the words.
The most well-known techniques include:
- Topic modeling: Identifying groups of words or collections of terms that best describe a main idea or theme.
- Sentiment Analysis or Document Classification: Determining whether a text is positive or negative. This analysis can be extended to any label that can be assigned to a text: classifying information received by a company while enabling spam filtering, grouping complaints based on their cause, analyzing the polarity of free-text fields filled out by customers, comments, and more.
Some Real-World Applications
Voice of the Customer
Understanding feedback from current and potential customers is crucial for any organization. The Voice of the Customer (VoC) refers to the ability to list and describe customer or user requirements for each company, including their perceptions and expectations about the product or service.
Today, the Net Promoter Score (NPS) is one of the most widely used indicators within a company to measure customer satisfaction and loyalty.
NPS is calculated based on the response to a single question, using a scale from 0 to 10: “How likely are you to recommend the product or service to a friend or family member?” It is a very simple metric to obtain:
- Those who respond with 9 or 10 points: promoters
- Those who respond with 7 or 8 points: passives
- Those who respond with 6 points or less: detractors
However, NPS by itself does not provide information that suggests immediate actions. Therefore, it is quite common for surveys to include a second question to help understand the reasons behind the customer’s rating. This question is usually open-ended: “Why did you give that rating?”, which requires much greater effort for processing and analysis.
Given the importance of such surveys, companies invest in conducting thousands of them, requiring teams of people to process and analyze the responses. In this case, processing time and costs can be reduced by using text analytics algorithms that help answer questions such as: What topics are most relevant to customers in my industry? What are the pain points and passions of our customers and those of the competition? among others.
Request Classification
Many customer service processes, especially in the public sector, are face-to-face and require a detailed interview between the service applicant and the organization’s representative. After the interview, the request must be classified and submitted for evaluation and/or prioritization. The classification process involves choosing labels (types) from dozens or hundreds of very specific categories, which requires significant time and effort from the person performing the task (usually the interviewer). In these cases, an automatic classification system that suggests, for example, the 5 most appropriate labels helps minimize human errors due to lack of knowledge of all categories, lack of experience, and/or the inherent subjectivity of any interviewer.
Incident Classification
Reading and classifying emails that arrive at a help desk service and creating tickets or work orders are tasks that consume a lot of employee time. Depending on the volume of messages received, these tasks can take up half of a person’s workday. However, these are automatable tasks where email classification algorithms play a very important role, as they directly help improve help desk efficiency and user satisfaction by reducing response times, thanks to alerts about urgency.
This application naturally extends to sectors such as insurance, where the main communication channel is still the telephone. In these cases, text analytics is applied to transcripts generated by a Speech-to-Text service.
Reviews and Brand Image
Using content from customer reviews or feedback at every touchpoint with the company to identify topics of concern in both positive and negative evaluations, pinpointing areas for improvement and/or the level of acceptance of new products and/or services.
Another direct application is Reputational Risk, which involves analyzing all comments made on social media about a brand or company. This option is used by organizations, for example, that operate in research or social support areas where there are no specific products and/or marketing campaigns.
Identifying Complaints
Customers use many channels to communicate with companies: emails, online forms, social media, and more.
When it comes to emails, often the complaint is only included in the body of the message. In other words, the customer does not indicate in the subject line that they are dissatisfied with the service, but rather details it in the email body, providing many specifics. If the email arrives in a mailbox that receives hundreds (or thousands) of emails per day, the company may fail to address the complaint in time and end up losing a customer. In this case, an alert generated by a text classification algorithm could radically change the time it takes to resolve the complaint and directly impact customer satisfaction.
Industries such as car rental or airlines with corporate clients (not just individuals) are examples where this text analytics application is highly relevant.
Sales Opportunities
In the international freight and delivery market, there is a system where participating companies communicate whether they have unfulfilled orders or available capacity to handle more shipments. This information is extremely valuable because scheduled trips that are not fully loaded can be optimized, making them more profitable. In other words, a company can benefit from another provider’s unfulfilled service if it can contact the end customer in time. This information is exchanged between providers via email, so if there is no timely alert, the business opportunity may be lost.
Another example is in the financial sector. Whenever the customer has agreed, within privacy limits, the institution can read the descriptions of direct debit receipts or credit card transactions to identify products and services contracted with third parties. With this information, the institution can design better products and/or determine the best timing for cross-selling campaigns.
Benefits for the Company
Some of the benefits derived from using text analytics models and algorithms include:
Cost and Speed: Compared to manual coding, the investment required for information processing is significantly lower. Additionally, Time to Value and Time to Market are reduced since conclusions are available in the short term.
Consistency and Uniformity: The same criteria are applied, avoiding inconsistencies that can occur in processes carried out by multiple analysts at different times.
Personalized Interpretation: Work is done with dictionaries, classification models, and sentiment vocabularies specific to the company. Ambiguity in terms is avoided, and sector-specific knowledge is incorporated.
Artificial Intelligence: Advanced Machine Learning techniques are used. We understand word importance, the distance to/among topics, and more.
Automated Procedures: Many tasks can be performed automatically after conducting a survey, receiving an email, etc. Other technologies such as Virtual Agents or Financial Advisors can also be used.
360° Customer View: Data from sources such as social media can be included to enrich the view of each customer. Additionally, new indicators can be created for use in other predictive models (such as churn propensity).