The definitive guide to text analytics for CX

A great deal of what’s said about your company comes in the form of text.

Whether it’s from reviews, support tickets, social media posts, or even blog posts, customers often engage by writing to, or about, your company. If you run surveys, you probably have plenty of open-ended text in your survey responses, too.

These are all great sources of feedback. They give your customers room to express themselves in whatever way feels most natural to them. Because you aren’t limiting their interactions with you by providing a predetermined set of options, it’s one of the most freeing ways customers can give you feedback.

But there’s one problem: How can you keep track of tens of thousands of individual pieces of feedback?

Many companies start by trying to read them all. This approach turns out to be a good one when you’ve got small volumes (less than 100 pieces of feedback a week). But as soon as you get into meaningful volumes of feedback, the time it takes to read and categorize all the pieces of feedback (and therefore the cost) quickly eats into all the commercial value you see from doing so.

It might work for a small company, but once you get to a certain scale, manual feedback categorization just doesn’t cut it.

That’s where text analytics comes in. And the benefits are many. Deploying text analytics at your organization is kind of like waving a magic wand. It's almost guaranteed to:

  • increase customer loyalty

  • Increase engagement

  • increase revenue

  • decrease churn

  • reduce the demand for your support team

  • optimize the management and handling of feedback

  • detect product issues

……. and the list goes on (and on, and on, and on). But, more on this later!

So what exactly is text analytics? Let’s go back to the very beginning.

Understanding text analytics

What is text analytics?

Text analytics is the process of using technology to extract information from large volumes of unstructured text. That text can come in any form or from any data source.

The idea behind text analytics is to feed open-ended text into a computer program that will help you to categorize, summarize, or otherwise analyze it to produce meaningful, quantitative insights. The information you get from text analytics can range from a list of keywords, an overview of the sentiment (e.g. if your customers are very happy or not) to a selection of the topics they’re talking about.

Rather than manually sitting down and reading each and every piece of text (a process known as qualitative data analysis), text analytics helps you identify the information that’s most important so you can take action, helping you consistently improve your product and customer experience—even if you’re a huge company.

Text analytics vs text mining vs Natural Language Processing (NLP)

You’ve probably heard of text mining and Natural Language Processing (NLP). These terms are often used interchangeably with text analytics because, at their core, they all refer to something similar—the process of extracting meaning from text.

There are some differences though:

  • NLP is when you use technology to process natural text, whether that is to understand the text or perform some sort of automated action like spam detection.

  • Text mining is a set of techniques to extract meaning from text. It includes things like counting word frequencies, clustering, stemming (separating the root of a word from a suffix or prefix), concept/entity extraction, categorization, and other techniques (including AI). Text analytics can and is used interchangeably with Text mining.

The difference in the definition between these three terms is subtle. What’s key here is to remember that NLP is not always concerned with extracting meaning from or understanding the text in a business context, which is what text mining and text analytics refers to.

Different approaches to qualitative data analysis

There are two key approaches to qualitative data analysis: top-down (or deductive) and bottom-up (or inductive)

What’s a top-down approach to qualitative data analysis?

A top-down approach is where the codes/categories/themes (the terminology is used interchangeably) are predetermined before your start the analysis. Your analysis then consists of reading through all the data to determine which codes apply to each piece of feedback you have collected.

A top-down approach to qualitative data analysis

This top-down approach is often used under three circumstances:

  • You already have an established set of categories you’re working with that you don’t want to change (perhaps you are using them to benchmark with old data).

  • If you’re looking for information about something specific, such as if your product team has just released a new feature and wants feedback about it.

  • You are using some sort of external research framework or methodology where the codes are fixed. This is much more likely in an academic setting.

But there are also a few disadvantages to this approach:

  • It’s inherently biased because it’s based exclusively on a set of predetermined codes. You’re specifically looking for things you want to measure even if it’s not really what’s most important in the data.

  • Only the things that you're aware of will surface in the data, meaning you’ll miss out on the chance to discover new themes.

What’s a bottom-up approach to qualitative data analysis

A bottom-up approach can be thought of as the more “data-driven” approach. It starts with an exploratory review of your data. You review the data (either manually, or with the help of technology and often just a representative sample) to inform what the appropriate codes/categories/themes are. From there, you go back through your data to code it, much like you do in the top-down approach.

A bottom-up approach to qualitative data analysis

The biggest advantages of this bottom-up approach are:

  • It’s great for recognizing emergent themes that might not have come up before or that your analysts didn’t know to look for.

  • It is the more “pure” approach from a data perspective because you are literally being led by the data.

The main disadvantages to a bottom-up text analysis are that:

  • It might miss issues that don’t occur that often.

  • It might fail to recognize that some terminology is related and assume an important issue isn’t that relevant.

  • It is very time-consuming if done manually.

  • Often, you will get slightly inconsistent results when done manually. Two people might come up with a different set of codes. The same person who does this analysis on the same set of data a few months apart is very likely to produce something slightly different.

Which approach is better?

There isn’t a one-size fits all answer to this question. The right approach depends on your specific use case. You need to consider a variety of factors, such as what you want the outcome you are looking for in the analysis, and what resources you have available.

The hybrid approach

Often, both styles of analysis are combined together for the same data to satisfy different outcomes (often referred to as the Hybrid approach). For example, when trying to answer a specific question about a piece of data, you might deploy a top-down approach, but if you are truly trying to “understand” some data, you opt for bottom-up.

Traditional text analytics has failed many CX teams because it often operates in a top-down manner where you or the vendor has to supply the code/categories/themes to use.

Accomplishing all three modes of analysis with one product is how you unlock the real power of text analytics.

Common techniques in text analytics for theme identification and classification

What techniques does text analytics use to both identify themes and code or classify those themes for you in your data? Generally speaking, there is one four techniques used to do this.

Rule-based classification

With this technique, you devise a set of “rules” to identify the codes/categories/themes you want to identify. The rules are things like the word “price” appears in the text or they use the phrase “great taste”. For instance, you might pre-select twenty key topics and their associated rules, feed this information to the text analytics tool, and measure how frequently those topics come up and what channels they’re being discussed on. This is one of the most rudimentary approaches to text analytics, but it is certainly more efficient than reading every piece of feedback at scale.

With these techniques, you have to specify the themes up front, as well as the rules to find them. These rules used to find or classify the themes are quite basic, which can lead to missing valid instances where the themes are being discussed that your rules failed to account for. There is also no way to capture new themes that appear in customer feedback over time if you don’t have a rule for them, which means you’re missing emergent issues as they arise for your customers (a classic shortcoming of a top-down analysis approach).

Supervised machine learning

This approach starts off very similar to the previous approach in that you need to specify the set of codes/categories/themes you want to identify. Once you do that, however, this is where it differs from the rules-based approach.

Not only do you need to supply your list of themes, but you also need to supply examples (ideally at least 500 per theme, but the more the better) where customers are talking about these themes. You then feed this to your supervised machine learning text analytics solution to “train” it on how to classify these themes in unseen customer feedback for you.

The training process is usually iterative and can take quite some time (usually measured in months). A simple way to think about this is the machine learning algorithm is developing its own “rules” in order to classify the themes for you (although in practice, they tend to be far more advanced than the rules we covered earlier).

As with the rules-based approach, there are drawbacks to this approach. Firstly, the training process is long and contains a number of manual steps. Secondly, just like the rules-based approach, you will miss any new themes that appear in the feedback. If you want to add them to your list of themes to classify, you will need to repeat the training process.

Finally, the reason why a piece of text has been tagged with a theme will not be apparent to you. The machine learning algorithm operates like a “black box” with no visibility into why it made the decision it has.

Unsupervised machine learning

With this technique, the text analytics program will use a technology called unsupervised machine learning to identify what it thinks the top keywords or themes are (depending on the particular algorithm used) in your text. A very basic explanation of unsupervised machine learning is that it is very good at spotting patterns in data.

Unlike supervised machine learning, you don’t need to train it what to look for. It will find patterns that it believes are statstically interesting and bring them to your attention.

This can be in the form of keywords, or phrases or it can even be a definition for a theme which comprises a list of weighted terms that it believes indicate the presence of the theme in the text.

The unsupervised approach is grounded in an old theory of linguistics first put forward by John Rupert Firth which states, “You shall know a word by the company it keeps”. Unlike rules-based and supervised machine learning, unsupervised machine learning is using a statistical approach to identify which words are correlated and identify patterns in how words are used in the data. Therefore, it is the only approach of the three that can support a bottom-up qualitative analysis approach.

This is all great news, however, there is a downside. There is usually very little recourse to make adjustments to the output of unsupervised machine learning algorithms (be it keywords, phrases, or themes). What this often means is that unsupervised machine learning by itself may not get you all the way there. Instead, it is left up to the user to decide how to use its output.

Hybrid machine learning

The last approach may well also be the most powerful. A hybrid approach combines together some combination of unsupervised machine learning and supervised machine learning and/or rules.

Hybrid machine learning often uses the output of an unsupervised machine learning algorithm to power a bottom-up qualitative analysis workflow and hence a shortcut to figuring out what the right set of code/categories/topics are, then use the power of rules and/or supervised machine learning to do the classification these themes.

The best solutions let you choose what is right for the job, be it rules or supervised machine learning. This approach is really the optimal approach when it comes to text analytics solutions because it can accommodate both top-down workflow (by manually specifying what themes or want to find) and bottom-up workflow (using unsupervised machine learning to figure out the themes as the input to a rules-based or supervised machine learning approach).

Additional texts analytics techniques for adding structure to text

Separate to identify and classifying codes/categories/themes, there is also a range of other techniques used to process data in its unstructured form and transform it into quantitative insights. Most of these are key parts of NLP and have varying degrees of usefulness largely dependent on the use case.

Some of the most commonly used techniques in text analytics are:

  • Sentiment analysis

  • Emotion detection

  • Intent detection

  • Keyword & Phrase extraction

  • Entity Extraction

Sentiment analysis

Sentiment analysis attempts to identify the tone conveyed in text data, which can be either positive, negative, or neutral. It helps you understand how your customers feel about a topic. Sentiment analysis can help you understand how satisfied or unsatisfied customers might be when talking about a topic and can be particularly useful for data sources that do not have a numerical score or rating, such as support conversations or social media comments.

As a simple example, consider this: If you’re a clothing retailer and customers have mentioned inventory-related topics 500 times in the past month, you’ll want to know whether that’s for a good or bad reason:

  • Customers could be positive - as in “there’s so much in stock!”

  • They could be neutral - “seemed to be a good number of options, but not my style”

  • Or customers could express negativity - “they need more inventory, everything was out of stock!”

Sentiment analysis detects and organizes this information in a quantifiable way to give you a clearer picture of how your customers actually feel.

Emotion detection

Emotion detection provides a greater level of specificity compared with sentiment analysis. Rather than the simplified, polarized options in sentiment analysis, emotion detection tries to parse out different emotions, like happiness, frustration, anger, and sadness.

Early on, emotion detection used lexicons, where certain words are associated with specific feelings. The difficulty is that many words can be ambiguous and meaning can often depend on the context. Slang—which varies by geography and subculture—can also be a barrier to accurate emotion detection. For example, the word “sick” usually has a negative connotation (“it made me sick”), but it can be slang for “great” in some English-speaking countries.

More recently, the focus has shifted to using the same Machine Learning techniques that you would use with Sentiment Analysis. However, it is a significantly more complex problem than Sentiment Analysis because it's reasonably common that a comment from a customer can have multiple emotions. For example, angry and surprised: "I can't believe how long I had to wait on hold, it was infuriating."

Contributing to the complexity is the sheer number of different emotions that could be expressed, leading to an increased probability that you can get it wrong.

In fact, at the time of writing, there isn’t even a real consensus on what set of emotions should be used. To illustrate this point, there is a 6-emotion model from Eckman and a whopping 27-emotion model from Google.

Intent detection

Intent detection is a text analytics technique that involves identifying the intention behind a customer's message or query. This is typically achieved by using supervised machine learning models to classify incoming messages into predefined categories such as "complaints," "billing inquiries," or "technical support."

Intent detection can be a powerful tool for CX professionals as it helps automate the process of routing customer requests to the right team or agent, improving response times and overall customer satisfaction. It also allows for the analysis of trends and patterns in customer intent, enabling businesses to proactively address issues before they escalate.

Keyword & Phrase extraction

Phrase extraction is a form of unsupervised machine learning in which—you guessed it—commonly used phrases are extracted from the text. This is a way to identify potential codes/categories/themes or to summarize the salient points in the text you’re analyzing.

The simpler form of this is keyword extraction, which works in much the same way. You can build a good picture of how it works if you imagine a word cloud. Word clouds display some words or phrases in larger fonts than others, based on how often they’re used. That’s a visual representation of phrase extraction.

Entity Extraction

Entity extraction involves identifying and extracting specific entities from unstructured text data. These entities can be defined as specific real-world objects, concepts, or individuals such as names of people, places, organizations, dates, and monetary values, among others.

The purpose of entity extraction is to enable the categorization, organization, and analysis of vast amounts of unstructured text data in a structured format.

There are various entity extraction algorithms and techniques, including rule-based systems, dictionary-based systems, and machine learning-based systems.

Rule-based systems rely on predefined sets of rules to extract entities, while dictionary-based systems use dictionaries of known entities. Machine learning-based systems, on the other hand, train models on annotated text data to automatically identify entities

What about Generative AI?

What is Generative AI?

Generative AI refers to a category of algorithms and machine learning models that can generate new data that resembles some set of training data (think ChatGPT or Bard). These models can learn from the patterns and structure in data, whether it's images, text, or any other form, and can then create new, synthesized instances that are similar yet unique. Generative AI is trained using a Hybrid approach, starting with the unsupervised training of a Large Language Model (like to Wikipedia)  before being further training using supervised machine to operate over a chat interface.

How Does Generative AI Work?

Generative AI models work by understanding and learning the underlying distributions of the data. They model the data in a way that allows them to generate new samples from that learned distribution. For example, Generative Adversarial Networks (GANs) are a popular type of generative model that consists of two networks: a generator and a discriminator. The generator creates new data instances, while the discriminator evaluates them. They are trained together in a game-like setup where the generator aims to create data that the discriminator can't distinguish from real data.

Is Generative AI Suitable for Text Analytics?

In short, no, or not yet anyway. Generative AI has shown remarkable results in various domains like art and image generation. However, when it comes to text analytics, especially in the context of CX professionals, it presents several challenges:

Lack of ability to validate: Unlike other models where the output can be compared to a known truth or benchmark, Generative AI models create information that doesn't have a corresponding reality. Trusting the results of Generative AI for text analytics is problematic because there is no way to validate what it is telling about the text is accurate short of doing a manual analysis yourself.

Non-deterministic Behavior: Generative models can exhibit non-deterministic behavior, meaning that they might produce different outputs with the same input. This variability introduces uncertainty in the analysis and could lead to inconsistent or contradictory insights.

Ethical Concerns: There's a risk that generative models can create misleading or fabricated information, leading to potential misinterpretation or misuse of data.

Relevance and Focus: Generative AI aims to produce new, creative content, whereas text analytics requires precise and critical extraction of insights from existing content. The focus on creativity might divert attention from the objective analysis needed for accurate CX understanding.

While Generative AI has brought innovative solutions to many areas, its application in text analytics for CX professionals is fraught with challenges. The inability to validate the information, the non-deterministic nature of the models, and other concerns make it a less suitable choice for professionals seeking to gain clear and reliable insights from their textual data. Careful consideration and alternative methods focused on deterministic and explainable results might be more beneficial for the field of text analytics in the CX domain.

Text analytics vs feedback analytics

Text analytics is one of the core features of a feedback analytics platform, but it’s not the same thing. Text analytics focuses exclusively on helping you uncover the meaning in unstructured and open-ended text. It’s one piece of the feedback analytics puzzle.

Feedback analytics software includes much more than that. It combines text analytics with data integrations, quantitative analysis, statistical analysis, segmentation analysis, trend detection, and reporting.

In other words, feedback analytics aims to build a broad, big-picture understanding of what your customers are saying, how that differs across segments & channels, and how it impacts the way customers behave across your entire organization—and text analytics is one of many tools involved.

The benefits of text analytics

Becoming data-driven has become a strategic goal for almost every business over the last few decades—with good reason (although I’d argue that becoming insights-driven is what they’re really after).

Data and insights broaden your perspective, allow you to question your preconceived notions, and enable you to make strategic decisions that really drive your business forward.

“But there are a million steps involved in collecting data, analyzing it, finding actionable insights, ensuring it’s accessible across your organization, and setting goals based on it.

A lot of companies find themselves stuck in the middle of this. They have an immediate problem and know the questions they want to answer, but they don’t have the infrastructure in place to find those answers when they need them. Either the data isn’t available or—more frequently—the data is right there, just not in a form that they can work with.

Text analytics helps you leverage large swathes of data that you’d probably never touch otherwise. And the benefits are immense:

  • Text analytics helps you process huge amounts of data, far beyond what you could realistically manage manually (even with a team of analysts). This makes it one of the most cost-effective ways to scale a customer insights program.

  • You can use it to make sense of unstructured text data. This is often data that most companies don’t leverage, despite the potential value of the information in there. Text analytics gives you a much deeper understanding of your customers and their needs, which you can feed back into your business to shape future decisions.

  • It makes your data more reliable and reproducible. Imagine you have a team of 1000 analysts who categorize your customer feedback. Training them and calibrating their categorization to produce a consistent output is a huge effort (and probably impossible, FWIW). An algorithm will always categorize things the same way. More reliable data means you can trust the conclusions you make more fully, leading to better and faster decision-making across your organization.

  • You can use it to generate truly actionable insights. Actionable insights are the bread and butter of any successful Voice of Customer (VoC) program. To get a return on your VoC investment, you need more than just interesting factoids. You need to uncover what you don’t know—unexpected connections—that enable you to make decisions that drive your business forward: by generating more revenue, acquiring more customers, and increasing customer loyalty.

  • Text analytics help you understand customer sentiment. Sentiment analysis can help predict churn, identify customers who need an intervention, or just make more sense of your social media mentions. It’s a powerful tool that should shape your brand choices.

  • It’s a way to listen more deeply to your customers. If you find yourself avoiding open-ended questions in your surveys, you’re missing out on an opportunity to hear direct, unfiltered feedback—which you can use to level up your customer experience.

Text analytics and CX

Using text analytics in CX

Everyone knows that building a great customer experience is a fantastic destination, but the challenge is how to get there.

Imagine being dropped in some foreign country—say Brazil—and being told you need to get to another city on the other side of the country by next month. The challenge is, you don’t speak Portuguese or have a map. Sure, you can start walking in a random direction, but are the odds actually good that you’ll achieve your goal?

Customer experience leaders often feel like this. Every company knows CX is important, but they don’t have the time, resources, or know-how to quantify how much of a difference CX makes in their growth trajectory. CX leaders feel like they have to navigate by gut instinct and hoping for the best.

It’s a strange dichotomy because while individual organizations often struggle with this, research has shown that investing in CX brings great results. For example, customers who have the best experiences spend 140% more on average than customers who have the worst experiences.

If you work with surveys like CSAT or NPS, you know how clear and simple KPI-driven reporting is. But you probably also know how challenging it is to move the needle on those KPIs without specific and accurate insights. You might already try to make ratings more useful by giving your customers a few, broad multiple-choice options for why they gave you that score. Not only are those options limiting, but it’s also likely that customers use different options when talking about the same issues (due to response bias).

Now say you’re in contact with hundreds or thousands of customers per week. Each of those interactions contains a wealth of valuable insights—if only you can figure out how to tap into it.

Text analytics helps you pull all of those scattered data points together into coherent, valuable insights that you can use to make better decisions.

Instead of being dropped into a country with no map or language skills, it’s like being dropped with a powerful personal translator in your pocket.

Ways to use text analytics in your business

When you start working with text analytics, you notice very quickly just how versatile it is. Because you generate text data almost every time you interact with customers, you can use text analytics in so many different areas of your business.

Let’s explore a few specific use cases.

Customer experience

Increase customer loyalty

Text analytics can help you understand the drivers of negative ratings, such as in your NPS and CSAT results. It can be part of understanding what the potential impact of making improvements can be. For example, Village Roadshow uses Kapiche’s text analytics to capitalize on the positive sentiment their guests had towards their events.

Reduce the demand for support

Text analytics can show you the main drivers of traffic to your help center and the chief reasons your customers are reaching out for support, without requiring you to manually tag or categorize every interaction. These learnings can supercharge your efforts to improve interactions with your product or service and grow your support organization more efficiently. Kind of similar to how CTU uses Kapiche for their students.

Prevent churn

Text analytics—particularly sentiment analysis—makes it easier for you to understand the emotions your customers are feeling. When a VIP customer starts to display frustration in their support tickets or survey responses, text analytics can quickly pick up on this so you can reach out to make things right. Rather than , text analytics enables you to resolve issues shortly after they appear. Don’t forget, it costs five times more to acquire a new customer vs retain an existing customer.

Improve customer experience

More insights about how your customers feel across their entire journey with you can enable you to focus your energy where it will matter most. Text analytics can proactively flag certain types of interactions, making sure you catch anything urgent or important.

For one compelling example of this, take Colorado Technical University (CTU). CTU used Kapiche to offer their students help within 24 hours when they mentioned they were struggling.

Optimize dealing with feedback

Often when customers leave feedback, they will be expecting something specific in response. For example, they may ask a question, threaten to leave your service or make a complaint with the expectation that someone actually reviews this and responds to them. Using intent analysis, text analytics can automatically detect cases of this for you and route it to the right people that can take action.

Product feedback

Increase customer engagement

By mining customer feedback on your product, you can understand why they use it, what value they see in it, and the scenarios that lead to frustration or decreased usage. Combining the output of your text analytics and other types of data—such as product usage data—can create a clearer picture of where you can improve.

For instance, perhaps you notice that you’re not converting trial users to paid users at the rate you’d hoped for. Combining this information with feedback from trial users—and perhaps segmenting based on whether they converted or not—can lead to important insights. Maybe your value proposition was confusing and your product doesn’t do what users hoped. Maybe you need to invest in an in-app onboarding experience. Or maybe your price point just seems too high.

Three very different conclusions, each requiring a very different action plan. The only way to know which route you need to take is to invest in text analytics so you’ll have deep customer knowledge at your fingertips.

Detect product issues

If your text analytics tool uses unsupervised machine learning, it can actually help you detect product issues you wouldn’t have known about otherwise. Because this approach transforms unstructured text into quantitative data—e.g. “X people are talking about Y negatively”—it can also highlight just how many customers are impacted by seemingly small issues. Nextdoor used text analytics to see that login issues dragged their overall NPS down by 2.3 points, then managed to drill down further to understand what those login issues were about so they could solve them.

Prioritize your product roadmap

The last thing you want is a product roadmap developed based on gut feeling. Yet, unfortunately, this is exactly how many product managers are forced to work. Data exists, but it’s hard to access and even harder to understand, so they’re forced to make guesses based on conversations with other employees and maybe a few customer conversations (both good things, but not exactly representative). If they’re lucky, they can then measure the impact of a product change on important KPIs after it ships.

Text analytics transforms the prioritization of your product roadmap because it enables you to understand what your customers are saying and feeling at scale. You’re not basing decisions on a few loud users or a vocal sales rep’s opinion. You’re making educated decisions based on insights and knowledge of your entire customer base.

Employee experience

Increase employee engagement

Employee engagement surveys aren’t all that helpful if they’re just a score. But when you can combine your engagement scores with emotion detection and categorization, you can understand the key drivers of positive and negative engagement. With text analytics, your people team doesn’t have to spend weeks or months analyzing the data. Run the survey, upload or push the data into your feedback analytics tool, and start processing results within minutes.

Text analytics unlocks your team to focus on improving the employee experience instead of spending countless hours reviewing survey responses.

Improve employee wellbeing

Burnout is a real thing, and it’s more prevalent than ever. Text analytics and emotion detection can be used to flag when your employees might be struggling, giving you a better chance of recognizing and reacting to potential burnout early.

When to invest in a text analytics feedback analytics solution

Investing in a text analytics solution becomes an obvious choice when you reach a certain scale.

Manually tagging and categorizing your text data can hold you over for a while when your volume is low, but it quickly becomes expensive and time-consuming. More importantly, it becomes a blocker that slows you down from making important decisions.

The tipping point that decides when it’s the right time to invest in a text analytics tool really depends on how much effort you’re investing into the analysis. If you have even one full-time analyst that’s spending multiple hours every day on manual work, it might be time to consider making a switch. An ROI calculator can help you understand the potential impact of investing in a feedback analytics tool for your business.

The real key here though, is understanding that text analytics is just one piece of the much broader feedback analytics puzzle. Text analytics alone won’t improve your understanding of your customers and skyrocket your VoC program.

It is when text analytics is used in conjunction with quantitative analysis, data integrations, statistical analysis, segmentation analysis, trend detection, and reporting that customer understanding is truly unlocked.

So how do I get started with Feedback Analytics?

Getting started with feedback analytics

Getting started with feedback analytics doesn’t need to be complicated.

Start by building a business case for it

Your business case for feedback analytics software should enable your company to see how much time and money can be saved. While it’s true that investing in feedback analytics can improve many different areas of your business, it’s often more compelling to focus on where you can save first. If you already have a Voice of Customer program or work with customer insights in your CX team, that’s a great starting point. Highlight how a modern feedback analytics tool will supercharge those efforts to give you even more ROI.

Acquire the tool(s) that will make it possible

Every feedback analytics tool out there has its own idiosyncrasies, so you need to know what questions to ask. When you know what your goals are, you can evaluate multiple tools based on them and see which one works best for your needs. One way to approach this might be to start with a set of questions that you want to answer with your data. You can shortlist a few tools and test them to compare the quality of insights you get from each one.
There are four key aspects you should consider when evaluating different tools:

  • The level of accuracy and insights

  • The time it takes to get results

  • How easy the output is to consume

  • Whether you can test it using your own data

There’s so much variety in text analytics software that it’s always best to see how it works, in practice, with your data before you make a decision.

Leverage the data in your organization

Feedback analytics is most powerful when you can combine data from across all of your data sources to create a full picture. Once you’ve pushed all available data into your tool, ask specific questions about the data to find the low-hanging fruit and build up momentum for larger projects. Make sure it’s accessible, share it when the opportunity presents itself, and see if you can create a regular report that’s compelling and actionable. Working with a new tool always takes a bit of time and some internal advocacy so that everyone knows it exists and what they could use it for.

Building a feedback analytics program is like turning a flywheel. It can take some effort up front, but once you get the ball rolling, it’s far easier to capture the momentum and tackle bigger initiatives.


Ultimately, text analytics (best served as a part of a broader feedback analytics platform) is a way to hear and understand what your customers’ needs and expectations are—so that you can actually meet them. The goal should be to deliver actionable insights quickly, without a ton of effort, and without taking huge amounts of time. Achieving this goal shifts you into a proactive approach where you can deliver insights quickly, provide valuable input when it’s most valuable, and drive your company’s future growth.

Kapiche was built with exactly that goal in mind. We use statistical, unsupervised text analytics within our feedback analytics platform and combine unstructured and structured data from any source you need, giving you the best insights possible—in no time at all. Check out a demo today!

Share to: