Obtain deep customer insights with MeaningCloud

Companies need to analyze the feedback that customers provide to them through a variety of unstructured channels: surveys, interviews, contact center, social media. However, the text analytics solutions available are limited to a shallow analysis of the feedback. In this post we show you how to use deep analytics to get a complete picture of customer opinions, perceptions, emotions and intentions.

Companies need to become customer-focused in order to understand the needs and opinions of their customers and thus, define the proverbial “segment of 1”. This forces us to implement Voice of Customer (VoC) analysis initiatives that go far beyond the typical periodic satisfaction survey with numerical scores, to look for new sources of insights.

Unstructured feedback: a great opportunity… missed

In this field, unstructured (which does not easily lend itself to being represented “in rows and columns”) and unsolicited feedback from clients is becoming increasingly abundant and important. Whether it may be fortunate or unfortunate, clients talk about us even when we do not ask them to nor want them to. The channels and means through which customers express themselves are varied:

  • Personal interviews, which are translated into audio recordings and transcripts.
  • Surveys, where open-ended questions are increasingly common so that customers can explain the reason for their ratings.
  • Contact centers, which are now intrinsically multi-channel and record interactions via voice, email, chat, bot and social.
  • Social media, with its full range of social networks, blogs, forums, review sites and communities.

Why is it valuable to analyze unstructured and unsolicited feedback? Among other things, because it allows us to circumvent some of the limitations of typical VoC techniques – which are based on conscious verbalization and recall – with their mistakes, biases and frequent tendency to give polite and insincere answers.

Unstructured and unsolicited feedback allows us to avoid these limitations because:

  • It is more spontaneous, sincere and it is expressed in the client’s own words, which allows us to understand how they perceive the problems and frustrations that they face.
  • It is more contextual – away from the artificial framework of many traditional techniques – which allows us to explore and discover new themes and relevant issues (after all, clients talk when and about what they want).
  • It has a viral capacity to get multiplied through the networks where it is published, which can create real reputation crises.

Nevertheless, extracting the value of  this feedback is not without its challenges, as it is an inherently big data problem: some companies may need to analyze millions of interactions per month (Volume), certain applications -e.g. social monitoring- demand very fast response times (Speed), and it deals with information that arrives in different formats: text, audio, image… (Variety).

Therefore, many companies lack the means to analyze unstructured feedback with the required parameters of volume, speed, variety and cost and this feedback is lost without being used.

Text analysis to the rescue… but with limitations

Text analytics, because of its ability to automatically convert unstructured text into structured data that can be used to extract insights, allows us to scale up the analysis of unstructured feedback. Text analytics technology is at the core of unstructured feedback analysis solutions in combination with  speech-to-text technologies (to transform contact center calls into analyzable text) or web crawling and scraping (to download content from websites, in media monitoring or market intelligence applications).

Text analytics technology incorporated in customer insight applications and tools is often limited in its capabilities and performs a very shallow analysis of feedback. In many cases, it is limited to tasks such as extracting entities, assigning categories at the document level and detecting general polarity.

So much so, that in a text like the one below:

“I’m tired of these TeleCom people. My cell phone keeps breaking down and I want to cancel the service. But their support center never answers. They’re the worst!”

the usual tools would probably not go beyond extracting the following data:

  • Entity: TeleCom (company)
  • General theme: Telecommunication services
  • General polarity: Negative

Thus, ignoring many elements of information that would enrich the analysis: What is the client’s mood? What is their intention? What attributes of the product/service do they dislike most?

Bridging the gap between data and value

We, therefore, need a new generation of text analytic tools that go beyond that and help us to turn results into Decisions, Decisions into Action, and Action into Value.

From a text analytics point of view, such value can come from a deeper extraction of the meaning embedded in comments, including insights such as the following:

  • Relevant Topics: so that the system can identify appearances of our brands, products, and companies.
  • Actionable categories: so that instead of generic categories, the system can classify the comment according to our functions, departments or organizations (for assignment and routing purposes).
  • Focused sentiment: a general polarity is not enough, we need to detect specific opinions about our products, their attributes and components.
  • Root causes: identify the ultimate, business-specific reasons for our customers’ problems or interactions
  • Drivers: understand the key levers and criteria governing satisfaction, quality or purchase
  • Discovery: possibility of detecting emerging issues, which arise in the conversation and which could be “off our radar”
  • Emotions: detect the emotional tone expressed by the client (joy, sadness, surprise…)
  • Customer journey: identify the stage in their journey and the client’s purpose: request information, evaluate products, buy, recommend.

In short, we need a 360º view of our clients’ opinions, perceptions, emotions and intentions.

How can MeaningCloud help you obtain deeper customer insights?

MeaningCloud allows you to perform a deep analysis of unstructured customer feedback, thanks to its solutions across three main axes:

  • Pre-elaborated Insights
  • Adaptation and customization
  • Development of new insights

Pre-elaborated Insights

Study the detailed opinion of customers with topic-oriented sentiment analysis

General, comment-level sentiment analysis can be useful when the text is short and talks about only one product or feature and expresses a single polarity (e.g. “I love the new iPhone”). However, when the text is long and talks about different products/companies, various attributes/characteristics within them and expresses different polarities for each one of them, a general, aggregated polarity can be misleading. Especially, given that it can compensate positive polarities with negative ones to provide an illusory neutral polarity and hide the different sentiments about the various polarity objects.

MeaningCloud’s Sentiment Analysis API provides, besides a general, comment-level polarity, a disaggregated polarity at the topic level, which are the entities and concepts that appear in the text. It can be customized with the incorporation of dictionaries, which include the topics most relevant to the analysis domain (e.g. brands and products and their characteristics). In this way we ensure that opinion analysis not only has a significant level of granularity, but also focuses on the aspects most relevant to our application. Try it here and learn more in this tutorial and recorded webinar.

Discover the emotional bond of customers with emotion recognition

The emotional bond customers have with our products is the hidden key to profitability. It is not only that emotions (especially negative ones) are durable, tend to be shared,  and can have a huge weight in shaping the customer experience, but that customers’ emotional motivators (how they want to feel) influence their behavior, especially in regards to purchasing and loyalty. Studies show that customers who are emotionally connected to our brand are (between 25% and 100%) more profitable than those who are simply satisfied with it. In fact, it is more profitable to invest in the emotional bond a customer has than it is to invest in their satisfaction.

How do we automatically detect our customers’ emotions? The Emotion Recognition Vertical Pack can identify the emotions expressed in a survey or contact center commentary. It is based on Robert Plutchik’s Wheel of Emotions, which is a standard model in the field and defines 8 main emotions: Joy, Confidence, Fear, Surprise, Sadness, Disgust, Anger, Anticipation. This analysis complements that of Sentiment so that, in addition to the granular polarity, we can obtain the “emotional color” of a commentary and thus, open up the possibility of a more complete management of the customer experience. Try it here.

Predict the future with intention analysis

What if we could predict the behavior of our customers on an industrial scale? The intention that our customers express in their comments (whether to buy, recommend, or abandon a supplier) is the factor with the most predictive capacity about their immediate behavior. Understanding their intention allows us to identify the stage of their “customer journey” that they are at and personalize the response to it. Also, discovering business opportunities (purchase signals), providing better service, retaining customers (churn prevention) and enhancing recommendation.

MeaningCloud’s Intention Analysis Vertical Pack identifies a series of basic intentions that may arise throughout the Customer Journey and can be explicitly expressed in a commentary: Information, Advice, Purchase, Support, Recommendation, Complaint, Cancellation. This way, signals are detected that allow us to prevent unwanted situations (Churn) or encourage desirable situations (Purchase, Recommendation). Try it here.

Evaluate the multidimensional perception of quality with the analysis of the voice of the customer

As is well known, the quality of a product lies in the perceptions of the customers, not in the eyes of the supplier. It is a multidimensional construct, which depends on aspects such as functionality, features, aesthetics, service and conformity to requirements. This perception of quality leads to customer satisfaction in both the short and long term.

The Voice of the Customer Analysis Vertical Pack allows us to evaluate customer perceptions through different dimensions both in a general context (independent of the industry) and specifically for different industries: Banking, Insurance, Retail, Telco… Its analysis models include a multidimensional hierarchy of evaluation axes and attributes, among which are the identification of the company, the interaction channel, customer service aspects, products, operations on them, or aspects regarding quality. In relation to the axes and attributes detected in a comment, the API tries to assign a disaggregated polarity. It also generates a measure of the general satisfaction expressed in the verbatim. In summary, this API offers an evaluation of customer perception through different dimensions and a measure of general satisfaction. Try it here and learn more in this tutorial and recorded webinar.

Discover the emerging topics and relationships between comments and text clustering

Just as important as making a detailed analysis of an individual comment according to predefined parameters is the analysis of a collection of comments which uses an exploratory approach to discover its implicit structure. Aggregating similar comments allows us to group them according to significant topics, identify relationships between groups and detect duplicates. Discovering the topics that emerge from the collection allows us to detect the “new voice” of the customer: comments that gravitate around entities or topics that we did not anticipate and were “off our radar”. These findings are a valuable source of business opportunities in the form of unmet needs or new product ideas.

MeaningCloud’s Text Clustering API allows you to aggregate similar texts and discover meaningful topics within a collection of comments. It applies unsupervised learning technologies and specialized text clustering algorithms to group comments based on their adherence to topics or content similarity, and thus discover the implicit structure of a collection of texts. Try it here.

Evaluate the collective opinion about your company through corporate reputation

Corporate reputation is a concept that evaluates public opinion based on a number of dimensions of the company: its leadership, innovation, products, financial situation, working environment, social responsibility… Although it is an intangible concept, many studies have shown that a good reputation increases the value of the company and provides sustainable competitive advantages. Traditionally, corporate reputation has been analyzed through studies and reports which use interviews and surveys to collect data (even with the limitations that these methods entails). Nowadays, companies want to have a more dynamic and immediate view of their reputation and to do so, they need to gauge the opinion in real time, on social media and in traditional media, as well as any other channel through which market feedback arrives.

MeaningCloud’s Corporate Reputation API is inspired by industry standards and performs an opinion analysis that combines topic extraction, multilevel theme classification and sentiment analysis. It identifies the organization that is being talked about, discriminates between the reputational axes and variables on which the comment or article is based and associates a polarity to each of them. The first level reputational axes that applies are:  Innovation and flexibility, Product and service offering, Financial situation, Strategy and leadership, Work environment, Integrity and transparency, Social responsibility. In this way, a multidimensional assessment of the company can be generated, obtained from all kinds of sources in real time. Try it here.

(IMPORTANT: The Corporate Reputation API is currently only available in Spanish. If you wish to use it in another language, please feel free to contact us at sales@meaningcloud.com.)

A much deeper analysis of unstructured feedback

The above pre-developed insights enable us to have a deep and immediate analysis of the unstructured feedback. For example, going back to the comment used at the beginning of this post:

“I’m tired of these TeleCom people. My cell phone keeps breaking down and I want to cancel the service. But their support center never answers. They’re the worst!”

many more information items are now available right out-of-the-box:

  • Entity: TeleCom (company)
  • General theme: Telecommunication services
  • General polarity: Negative
  • Emotion: Annoyance
  • Product: Cellular phone
  • Attribute – Reliability: Negative
  • Intention: Cancellation
  • Service – Support: Negative
  • Overall Satisfaction: Negative

In short, these pre-elaborated solutions allow us to extract a deeper meaning from unstructured comments.

Adaptation and customization

Adapting text analytics to the application in question is the key to achieving accurate, actionable and valuable results:

  1. From the point of view of information extraction, it is necessary to be able to detect mentions of our company, brands, products, relevant attributes, competitors, etc.
  2. From the point of view of classification, it is essential to define categorization models that reflect the root cause of problems or interactions that our customers have, the drivers of satisfaction and purchase criteria, the functions/organization of our company (for the purposes of routing and allocation), etc.
  3. From the point of view of opinion analysis, it is very useful to be able to define the polarity that certain expressions entail in certain contexts. For example, the phrase “It has the highest interest rate in the market!” is positive, if it talks about bank deposits, but negative, if it talks about mortgages.

MeaningCloud’s customization tools allow you to create dictionaries, general classification models, deep categorization models, and custom sentiment models to tailor the operation of text analytics engines to your application for maximum accuracy and relevance. This is done with user tools that guide you through the development of these customizations without the need for programming.

Development of new insights

Imagination is the limit when it comes to developing insights that add value to customer understanding. New ways to represent and evaluate the perceptions and desires of our clients can help us make decisions and turn unstructured information into an advantage.

It does not matter if those new insights are not currently implemented in our text analytics product. MeaningCloud, thanks to its Professional Services offering, can collaborate with you in that development. These are some of the new insights we are implementing for some of our current clients:

  • Brand associations: What entities/concepts do people usually mention when they talk about our brand? Which features are the most mentioned? Do they talk about us in relation to one product category or another? This analysis is equivalent to discovering a kind of “semantic footprint” of our brand and provides an individual, high-granularity perceptual analysis.
  • Perception analysis: It allows us to understand how our clients perceive our brand, with respect to certain relevant pre-defined attributes, and compared to competitors. This data allows us to generate an aggregated competitive perceptual map of our sector (a classic marketing tool) that is the basis for the analysis of our brand positioning.
  • Brand personality: If our brand were a person, what kind of person would it be? Personality analysis allows us to identify the human characteristics attributed to a brand: Sincerity, Excitement, Competence, Sophistication, Ruggedness… There are accepted models for brand personality analysis, such as Jennifer Aaker’s.

Are you interested in having us develop these or other insights for you? Our Professional Services department will be happy to help. Please do not hesitate to contact us at sales@meaningcloud.com.

Conclusion

Unstructured customer feedback is more valuable than we might think, because of its enormous potential to uncover customers’ opinions, perceptions, emotions and intentions. However, in order to exploit unstructured customer feedback with the necessary quality, volume, speed and cost requirements, we need a new generation of text analytics tools. Fortunately, the right tools to extract such value are now starting to become available.

Learn more, including demos, in our recorded webinar.


Leave a Reply

Your email address will not be published. Required fields are marked *

*
*