In this guide, we'll teach you the ins and outs of customer sentiment analysis so you can turn surveys, reviews, social and support conversations into actionable customer insight.
In short, a customer sentiment analysis will show you how your customers are reacting, positively or negatively, to your company’s products, features, and processes.
When you read a piece of a customer feedback—let’s say a customer review—you can immediately pick up on whether that customer is happy, unhappy, or...just neutral. You’ll likely also be able to understand why they’re feeling the way they are.
That’s all a customer sentiment analysis is: detecting the sentiment of a customer from text feedback.
The applications of customer sentiment analysis are numerous. But the core benefit to more clearly understanding your customer feedback is to identify pain points, which can then be fixed to drive customer growth, loyalty and retention.
Customer sentiment analyses can (and probably should be) automated. Most companies have thousands of reviews, survey forms and customer support conversations happening every month.
An automated sentiment analytics tool allows you to stay on top of your customer’s ever-evolving frustrations and pain points, without the considerable manual outlay that comes with a member of your team reading -> analyzing -> understanding -> tagging every single line of feedback.
Sentiment analysis models can focus on polarity, emotions, urgency and intentions.
Depending on the results and use case you want to achieve, you can tailor your sentiment analysis approach.
Most automated approaches will use complex dictionaries of words that are typically interpreted as emotions, intentions, positivity and negativity.
For example, if a customer leaves a Facebook comment that says “your order process is so complicated”, a sentiment analytics tool should automatically pick up the aspect ‘order process’ and the sentiment, e.g. ‘complicated’ is usually interpreted as negative and frustrating.
The approach on the previous example is called ‘aspect-based sentiment analysis’, it’s designed to identify both emotions and the ‘why’ behind them.
At scale, say across 50,000 customer support conversations or customer complaints each month, this can be an incredible useful way for a brand to get a real-time understanding of their customer’s positive and negative company interactions.
To illustrate the point, let me start with these three stats:
Source: CX statistics
You can probably see where we’re going with this…continuously conducting customer sentiment analysis is critical because it allows you to understand your customer experience better. Only then, of course, can you improve it and reap the rewards.
Ultimately, a customer sentiment analysis tool is your channel to better customer experience.
But, how does it do that?
A manual approach to sentiment analysis is fine when you only have a handful of pieces of customer feedback to analyze.
However, and we’ve touched on this before, when you have thousands of comments, reviews, surveys, and support conversations on multiple channels and platforms, a sentiment analysis tool allows you to scale your analysis and cover every data point rapidly.
Many overcome this challenge by taking a small sample of randomly selected pieces of customer feedback. This is one option, however, you’re likely to miss crucial context and nuance that is persuasive across your company.
There’s nothing more impactful than going to your website team and saying “10% of all customer feedback is for the payment portal and 90% of it is ‘extremely negative’”.
The speed of sentiment analysis enables a real-time understanding of customers. Identifying issues as they unfold can be extremely powerful for improving CX and preventing churn.
For example, with real-time sentiment alerts you can know immediately when a negative PR story is unfolding on social media, or a particular batch of products was damaged, allowing you to take a preventative approach to handling it.
For your analysis to be ‘actionable’, meaning other teams can (and want to) rely on the results to inform their projects and roadmaps, it must be objective.
Human beings are naturally biased, they understand the world through their own tainted lens. For analytics purposes, this means that the outcomes of a manual customer sentiment analysis may be different depending on who conducts it. How can other teams in your company be sure they trust it?
Automation quickly overcomes this and applies one consistent lens to all your customer feedback.
Related read: What are actionable customer insights?
These three points, scalability, real-time and objectivity are beneficial beyond the customer experience department.
The results of a sentiment analysis have a ton of useful benefits for multiple teams in your organisation (as you can read in the article, teams like product, operations and marketing can leverage them to hit their KPIs, too.)
Before we get stuck into the step-by-step process of sentiment analytics, it’s important to note that it’s time-consuming and subjective when done manually.
This method will work well only with a small number of feedback pieces, a lot of time, and just one person doing the analysis—any more may change the results because everyone interprets feedback differently.
For larger volumes of customer free text feedback, it’s simple, affordable and provides significantly better results to automate the analytics process. Explore our product—which analyses surveys, reviews and all customer support channels in real-time—and book a demo with us here.
For those who endeavour to go on, let's get stuck in.
For this manual analysis, we’re going to do an easy aspect-based sentiment analysis.
Support conversations are the most useful data source for customer feedback. Support ticket logs (from emails, calls and live chats) contain unbiased, qualitative feedback that's an unbeatable source of customer insight. We've written extensively about why support is the most valuable source of insight, here's one such example on support ticket insight.
You should be able to export your chat or email logs for this exercise, and then take a small sample to familiarise yourself with it.
Reviews are a core driver of sales as most customers look there before making a purchase. This makes understanding the reasons behind negative reviews important for business growth.
P.s. British Airways currently uses our sentiment analytics tool to analyse all their customer reviews, read the case study to see what they are achieving.
NPS and CSAT are the most common ways to collect customer feedback. Understanding the drivers behind positive and negative sentiment at scale is a powerful use case for AI-powered sentiment analytics. However, NPS surveys are now considered a biased form of feedback, so you'll need to work hard to ensure your results are actionable.
Export your feedback and collect all the raw data in one place (e.g. an Excel sheet). With it include any extra information that might help later on like...where did the feedback come from? Who is the customer? What CRM data can we pull on them?
You’ll want to collect the same data on each piece of customer feedback to have a complete dataset and a meaningful result ultimately.
In the template we’ve used below, we decided to include:
Once your customer feedback data set is in one place, you need to think about how you’re going to categorise the data.
You’ll need two spreadsheets. One for the feedback you’ve already collated, and another to store the labels with which you’ll code the feedback.
Here, we’ve created a very basic starting point for you to build upon.
For our feedback dataset—Trustpilot reviews—we decided to use a simple 1-5 polarity labeling system. There are a number of ways to do this, our engineers use insanely complex automated labeling practices to speed up the process and ensure insights are granular and actionable.
You can expand these rules as much as you like, the more detailed you can go the more complex your analysis will be but also the more actionable the results will be.
You can also go beyond polarity labelling to more complex emotions, urgency and intentions to get more from your analysis. One favoured tag by our customers is the “ChurnRisk” tag which helps us to identify in real-time when a customer is experiencing a problem or emotion that will likely lead them to cancelling their account or leaving for a competitor—making prioritization of those customers fast and simple.
Here’s an example of our basic topic taxonomy and sentiment analysis applied to one survey result. You’ll achieve the best results by setting up a topic taxonomy of topics in advance so that the aspects (the positive and negative topics) are uniform in the end result.
We interviewed a number of experts to write a guide to building a tagging taxonomy here.
How you report a customer sentiment analysis is almost as important as the analysis itself.
Your sentiment analysis report needs to be not only understood by others but acted upon. The customer feedback you present should strike a chord with the audience emotionally, while filling them with confidence that they can take this feedback seriously.
Read our eBook here to see how industry experts like yourself are making it easy to understand the value of CX data.
Let’s also take a look at some examples of how you can report your customer sentiment analysis results.
These are taken from the SentiSum platform:
If you’ve done an aspect-based sentiment analysis like you can with the SentiSum analytics platform, you can easily present the topics along with how customers feel about those topics.
If you’re working on a new project to improve a particular feature, a report on changing sentiment can help you track performance.
Furthermore, this report can help your team identify dramatic swings in sentiment that could do with a root cause analysis.
This graph shows how customer sentiment has changed over time for a topic, in this case ‘shopping experience’.
It’s helpful at a high level to understand how customer sentiment is improving for a topic and a visual graph shows clear progress to your team.
Here’s how we present it. Lots of visuals, time series, patterns, ability to deep-dive. Those are some ideas, you can attempt within google sheets.
In the example below, taken from the Sentisum deep-dive view, one of our customers is able to report to their team which topics are driving customer support contact and in what volume.
You can use help statistics like % of Contact Per Order (CPO) to give context to your colleagues. In this example, customers contacted support about ‘recipe feedback’ nearly 10,000 times, which contributes 0.87% of their overall CPO.
You might want to include a core statistic you’re tracking, or a more simple one like ‘what % of overall customer feedback is about this topic?’.
It’s useful to know that in your survey ‘payments’ were mentioned 100 times, but if you started with 10,000 surveys then 100 mentions of this topic makes up just 1% of the issues mentioned.
We built a very simple customer sentiment analysis template to get you started.
Download it for free below 👇
When leveraging the template created for this article, you should start by reaching out to colleagues across your company.
Every team has different interests and needs from a sentiment analysis, and with their help you can pre-plan so the results are genuinely useful across your company.
If other departments are involved early, and their goals are taken into account, they’re also much more likely to listen to and take action on your final analysis findings. It's a win-win.
Download it here 👇
As we've mentioned a number of times, sentiment analysis is really something you want automated. It's time-consuming and subjective to do manually, so any meaningful analysis will come from an automation tool.
There is a wealth of customer sentiment analysis tools to choose from, which can make it difficult to choose.
We suggest booking a number of product demos to look for a tool that can help you with your problem—you’ll get a good insight into the company and how they work through their sales process.
A company you want on your side will invest time in teaching you about the market and solutions available.
When choosing a tool, we suggest asking yourself these three questions:
SentiSum uses AI technology to completely automate your sentiment analytics at a root cause level. We specialise in customer support conversations, but have solutions across NPS, surveys and almost all feedback channels. Learn more here.
Talkwalker's quick search looks at your mentions, comments, engagements, and other data to provide your team with an extensive breakdown of how customers are responding to your social media activity.
Businesses use Brandwatch to monitor mentions online and understand the voice of customer, detect fluctuations in sentiment, and measure brand visibility in real time, 24/7
Perform multilingual sentiment analysis using MeaningCloud. This online tool runs aspect-based sentiment analysis to decide whether specific topics are mentioned in a positive, negative, or neutral way.
Rosette has an API that uses AI to analyse natural language. They've branched out from social media to analyse entire documents, for example, the sentiment expressed by customers when they mention a specific product, company, or person.