Your team spends hours on end trying to come up with the best survey to gain a deeper understanding of customer experience and the areas it needs to improve. You sign up to SurveyMonkey and create professional surveys in a matter of minutes. Done!
Then, your team receives the survey responses and can easily analyze most of them… until they reach the open-ended responses. What is the customer saying about UI? Has the overall user experience improved? These questions need to be answered so you can create reports and analyses (asap).
You brace yourself for hours of reading and manually categorizing the written responses, to find insights that will help answer your questions. You finish going through the responses and send the report just on time. Then, a new batch of survey responses arrive... Time to go through this emotional rollercoaster all over again! Well, not necessarily.
There’s no escaping the fact that we need to understand the Voice of Customer (VoC), and we try our best to do so. That’s why more than 40% of SurveyMonkey’s surveys include open-ended responses. However, it’s hard to grasp the insights that lie behind heaps of text. After all, how can we measure words? Well, one effective and easy way to do it is with text analysis.
A lot can be done with text analysis, but you might be wondering won’t it take a while? You’ll be glad to hear that the answer is ‘no’. You don’t need to invest a considerable amount of time or know how to code. You can get started right away and here you’ll learn how!
In this post, you’ll find the following information that will help you get started with automating the analysis of your surveys in SurveyMonkey:
Topic detection, a type of text classification, is the process of automatically tagging (assigning categories or tags) according to the content of the text. These categories must be pre-defined according to your needs, so if you want to know about Price, Design, and UX, you can use tags for these topics and tag them accordingly.
You can also use sentiment analysis, another type of text classification that focuses on classifying opinions with positive, neutral, and negative tags.
You can go even further and combine topic detection with sentiment analysis to create what is known as aspect-based sentiment analysis. This technique not only helps you learn what people are talking about but also how they are talking about it. That means you can understand, for instance, how positively Price, Design and UX are viewed by your survey respondents.
Keyword extraction is the process of identifying the most relevant words or expressions within a text. So, if you want to understand what’s going wrong with your product, you can check the words — keywords — that frequently appear in the open-ended responses. Consequently, you might learn that negative reviews regarding UX are related to loading time, bugs or difficult to use.
What do our clients want? This is a question that businesses have asked themselves time and time again, and sometimes they don’t have an answer. That’s why we conduct customer surveys, after all. But nowadays clients expect more from the products and services they buy, so much so that according to a survey conducted by BloomReach, 87% of 2,000 clients said they’d rather buy from a company that can predict their intents and needs.
So, we can all agree that we need to analyze open-ended responses from customer surveys, in a way that provides key insights as quickly as possible. And text analysis with machine learning is the answer since it can deliver:
Most businesses don’t have time to spend hours reading, tagging and re-reading hundreds of survey responses to get insights. It’s also expensive since you need to hire extra agents just to help manage increasing workloads. On the other hand, text analysis can turn hours, days and weeks into mere minutes, saving your team valuable time so they can focus on more important and fulfilling tasks.
When survey responses are negative, sometimes it’s key to pinpoint where the problem lies to solve it straight away and avoid a customer from churning. This is easier said than done with manual labor, but with text analysis you can get the latest survey response insights in real-time and make key decisions right away to prevent customer drop-off.
We try to follow guidelines, but sometimes we have doubts, or get tired when manually tagging survey responses. Who wouldn’t? To make sure insights are accurate, we need precise and constant criteria, and for that, a centralized text analysis system is key.
Thankfully, analyzing survey responses automatically with text analysis is fairly simple, and after following our tutorial, you’ll be able to create your own model that’s able to provide accurate results tailored to your needs.
To start, you have to get the responses that you want to use as examples to train your machine learning model.
SurveyMonkey provides an easy way to export survey responses by clicking on ‘Analyze Results’ in the menu bar:
Then, select ‘Individual Responses’ and choose the open-ended responses you want to download:
Finally, you’ll have to select the answers to the open-ended question you wish to download, click on ‘Export’ and choose the desired format. Keep in mind that, for analyzing the survey responses with MonkeyLearn, you’ll need an XLS or CSV file format.
Alternatively, you can integrate SurveyMonkey with Google Sheets to easily export the answers to a spreadsheet:
After installing the add-on, a new window will appear within Google Sheets. Now, you can select the survey responses you want to appear on your spreadsheet, and you’re good to go. You can then download it easily as a CSV or Excel file:
Another tool that can come in handy is Zapier, but we’ll cover that in more detail below.
Once you’ve gathered your survey responses in a CSV or Excel file, you can create your very own sentiment analysis model that — once it’s been trained with your data — can analyze the responses automatically for you.
In the following tutorial, you’ll learn all you need to know to create a sentiment analysis model for your SurveyMonkey responses in just a few simple steps:
1. Create your model
After creating your MonkeyLearn account, go to the dashboard and select ‘Create Model’. These options will appear:
Since sentiment analysis is a type of classifier, you should click on ‘Classifier’ and then on ‘Sentiment Analysis’:
2. Import your SurveyMonkey survey responses
Now, it’s time to select your source of training data. Since we’re using the SurveyMonkey responses we just exported, it’ll most certainly be a CSV or Excel file. So click on the file format in which you downloaded the responses, select the file and upload it:
3. Begin the training!
It’s time to teach your model to predict positive, neutral, and negative responses. You’ll need to do a bit of manual work to train your model, but once it’s ready, you’ll save hours in the long run, so it’s definitely worth it.
As each response appears, tag it as Positive, Neutral or Negative. This way the model will learn your criteria and identify patterns from the examples you provide to make future predictions:
After a few minutes, you’ll start noticing that your model begins to make predictions that you can either confirm or correct. Not quite accurate yet? Don’t worry. This just means it needs more training. The more data you feed it, the more accurate it will become.
4. Test your model!
Once you’ve taught your model how to tag different responses, it’s time to see if it’s ready to roll! Just go to the ‘Run’ tab and have a go at writing your own SurveyMonkey response:
Your model may still need some work if it’s not making the right predictions. But that’s okay! Just go back to the ‘Build’ tab and keep tagging some more examples.
Also, keep in mind that you can check the model’s accuracy by clicking on ‘Build’ > ‘Stats’ where you’ll be able to see the F1 score and the accuracy of your model:
You can check the precision and recall for positive, negative or neutral tags:
However, you’ll need to have a minimum of four texts per tag to be able to see these stats. If you want to know the meaning of F1 score and recall, among others, you’ll find the explanation here.
The keywords that appear in the image above are those that appear more often in texts that are tagged as Positive. This is called the keyword cloud and it’s a good way to quickly see what customers mention most in each of the categories: Positive, Neutral, and Negative.
Is your model still making inaccurate predictions? Go to ‘Build’ > ’Data’ to see if there are any false positives or negatives (incorrect tags) that could be confusing your model and. If that’s the case, assign the correct tag to the data and get your model back on track.
5. Put your model to work!
Yes, it’s time to upload new survey responses that you want to analyze and get predictions for! To do this, you have three options:
a) Go to ‘Run > ‘Batch’ and click on ‘New Batch’ to upload a CSV or Excel file to run a batch analysis with your sentiment analysis model. Select the file you want to analyze and then receive your sentiment analysis predictions, which you can download to your computer in seconds:
b) Use one of the available integrations. Perhaps Google Sheets if you have the responses in a spreadsheet. Just click on ‘Integrate’ and select your integration:
c) Last but not least, if you know how to code, you can make use of MonkeyLearn’s API:
Do you want to find out what your clients are talking about? Then a topic classifier is the tool. The process is quite similar to that of a sentiment analysis model, but there are some differences.
The first difference you’ll come across is the type of classification you’ll want to do. Instead of clicking on ‘Sentiment Analysis’, go for ‘Topic Classification’.
Next, you’ll need to upload the training data. Now, here’s where a topic classifier differs slightly to a sentiment analysis classifier – you’ll need to define your tags, that is, the topics that you want to get insights for. For example:
One thing to take into account is that the more tags you have, the more data your model will need for training. Once you’ve defined your tags, you can follow the same steps we outlined for the sentiment analysis model.
Once your model starts making the right predictions, you can analyze a batch of SurveyMonkey survey responses by going to the ‘Run’ tab and selecting ‘Batch’ to upload an Excel or CSV file, clicking on one of the available integrations or the API.
Moreover, if you’re hoping to get more in-depth knowledge from your survey responses, you can also run an aspect-based sentiment analysis to understand what customers are talking about in their responses (topic) but also how they feel about them (sentiment).
First, analyze a batch of survey responses with your sentiment analysis model and then use a topic model to analyze the same file. In the end, you’ll have a file with each response, its sentiment prediction in one column and the topic in another.
Imagine you’ve already analyzed your responses with a topic classifier, and the dev team wants to know what customers are talking about negatively under the product tag… To do this in the fastest way possible, you can use a keyword extractor. This machine learning model can detect the most relevant words or expressions from survey responses, things such as bug, issue and slow, and give the team an idea of where they need to concentrate their efforts to improve the product.
Take into account that ‘keywords’ are not going to be the same for every team and business. What you consider a keyword may vary depending on your team. For example, if you’re in sales, you may consider billing, price, paid plans and discount as keywords, but UX and design? Not so much. That’s why it’s a good idea to learn how to create your own custom extraction model. However, if you’re not there yet, you can use this pre-trained keyword extractor right away.
You’ve trained your text analysis models and they’re running smoothly so you can focus your energy on more pressing tasks… but you still need to upload and download batches to for text analysis. Is there a way to save time on this repetitive task of click, drag and drop? Well, there’s a tool called Zapier.
So what is it? Zapier is a very user-friendly tool that automatically integrates actions and data from different apps without you acting as the middleman. It’s based on two concepts:
By way of triggers and actions, data passes back and forth between different apps. Each workflow with triggers and actions between different apps is called a Zap.
If you have a SurveyMonkey annual plan – which allows access to SurveyMonkey answers in Zapier – you can automate the analysis of SurveyMonkey responses with MonkeyLearn in just a few simple steps:
Ok, you get the idea. But let’s see how you could create this integration so that everything runs seamlessly in the background (ah, that sounds nice). Because Zapier has an integration for MonkeyLearn and SurveyMonkey, this step-by-step process will be a piece of cake:
1. Log in to Zapier
2. Start creating your Zap
On your dashboard, you’ll spot a flashy orange button that says ‘Make a Zap!’. Click on it!
3. Define SurveyMonkey as your trigger
Time to set the trigger app: SurveyMonkey and the specific trigger, an incoming survey response from SurveyMonkey. Just click on ‘New Response Notification with Answers’:
Don’t forget to click on ‘Connect to Account’ to get your surveys from SurveyMonkey.
4. Define the action you want MonkeyLearn to do
We already have our trigger. It’s time to set our first action. Select ‘Add a step’ and click on MonkeyLearn as your action app. Then, select either ‘Extract text’ or ‘Classify text’ —depending on the text analysis model you’ve built — as the specific action. Remember to click on ‘Connect to Account’ and paste your MonkeyLearn API Key:
Now, you’ll have to select the model you want to use. In this case, we selected ‘Sentiment Analysis’. Next, select the trigger action (SurveyMonkey survey response) as the text to analyze:
As mentioned earlier, take into account that selecting the survey responses in Zapier is limited to SurveyMonkey premium users.
If you decide to run an aspect-based sentiment analysis on your survey responses, you’ll need to add a second action to your Zap: a topic classification analysis.
5. Define the action to get the results on a Google spreadsheet
To see the analysis results on Google Sheets, it will be necessary to add a final step to your Zap. Select ‘Add a step’ and Google Sheets as the app. Then, you’ll have to select the type of action you want to do:
After selecting the Google Sheets action, connect your account. Once that’s done, click on ‘Continue’ and pick the spreadsheet you want to use to receive the SurveyMonkey text analysis results:
Finally, click on ‘Continue’.
6. Turn on your new Zap and give it a go!
Click on ‘Finish’, name your Zap, and turn it on! Voilà! Your text analysis results will appear automatically on your spreadsheet without any copying or pasting!
Now that you have a fast and effective way to analyze your survey responses, you can share them with your team. You’ve noticed that there are some insights worth looking into and you need your team to be on the same page to begin with the decision-making process.
You share the spreadsheet of responses alongside the analysis results and… silence. Your team members take a minute to read and compare the predictions… then a few minutes more. How can you gain efficiency in a meeting and convey the findings? Well, visual language shortens meetings by 24%.
Besides increasing productivity, visualizing data in beautiful graphs and charts helps clarify complex information within seconds so that team members can grasp the insights easily and feel empowered to start pitching ideas.
If you want to try presenting a graph in your next meeting, there are various data visualization tools that are very easy to use. Below, we will introduce the most popular ones so you can start seeing and stop reading results.
Got your results on Google Sheets? Then it’ll be easy to integrate to your Google Data Studio account. Just click on ‘Data Sources’ on the left side of the screen and then on the blue plus button at the bottom right-hand corner of the screen:
A list of available data sources will appear, so look for Google Sheets. Now you’ll have to select the spreadsheet you want to connect to your report in Google Data Studio.
Keep in mind that if you have your text analysis data in a CSV file, you can upload it to Google Cloud Storage, which is one of the available data sources, and then select it as a connector for your report.
Once you upload your data, it’s super easy to create beautiful charts with Google Data Studio. You can create tables with the most relevant data by clicking on ‘Table’ in the toolbar, or selecting ‘Insert’ > ’Bar chart’. If you want to style your report, you can go to the ‘File’ section, click on ‘Report and theme settings’ and select the colors you want for the background and the bars. To finish, add a banner with the title of the report and share the data with your team.
To learn more about creating amazing graphs with Google Data Studio, go through this easy step-by-step guide.
With this data visualization tool, you can create attractive reports (called dashboards) with different charts (called tiles).
One of Looker’s most interesting features is the filtering: if you have a graph for aspect-based sentiment analysis of your SurveyMonkey responses, you can pick one of your topic classifier tags, for example, Price, and take a closer look to see what people think about the price of your product.
Although Looker currently doesn’t have an integration with Google Sheets, you can use a simple script to pull information from a Google Sheet and add the data to a database that is integrated with Looker!
Want to learn more? Check out some of Looker’s tutorials on Youtube.
In just a few seconds, you can upload your survey results in a CSV or Excel file to Tableau, although that’s not your only option. There are many databases you can connect to. Once that’s done, all that’s left is to create your graphs by dragging and dropping data. Before you make up your mind, you may want to take a look at this tutorial.
If these tools don’t feel quite right, other apps may have what you’re looking for. One popular app is Mode Analytics, which comes with a very convenient Slack integration for teams that communicate with this app. Klipfolio may also be a good call but it’s a bit complicated if you’re hoping to upload your data from an Excel file. It’s best for cloud integrations.
There’s no denying that you’ll save valuable hours by implementing text analysis on your survey responses, but how will it impact your business? Let’s take a look at some of the advantages:
Seeing churn on the rise is a nightmare for teams. A big portion of your churn rate might be down to poor customer experience. For example, customers might express dissatisfaction and get no response… because you were busy analyzing all of the survey responses manually. Meaning their issue was never identified or resolved. Not anymore.
Now you can keep track of dissatisfied customers that write, for example, ‘I couldn’t log in all day, there must be something wrong…’ without delay by checking the text analysis results. Then, you can take the necessary steps to promptly solve the problem and offer the client some sort of compensation.
Clicking, scrolling and clicking again to track down a response that concerns your team can be quite repetitive and inefficient. Instead of going through the motions, you can save your team valuable hours using topic classification. Just create a model with tags for each team and the survey responses will be assigned directly to the team in charge.
Imagine you get a SurveyMonkey response tagged under Product or Bug. In this case, the response will be routed to the dev team. That way, you’ll not only be able to send the ticket directly to the right team that’s able to solve the problem, you’ll keep the customer happy by providing a fast response. Plus, you’ll also prevent future bad experiences that slow down the routing process..
Maybe you’re facing a dramatic decrease in sales or an unexpected influx of tickets... but this shouldn’t be unexpected. Understanding the Voice of Customer can help you predict trends and grasp where you’re going amiss.
By making use of an aspect-based sentiment analysis model, your team can notice how high the number of negative responses is and, furthermore, understand what aspect of the product is having a negative impact on the customer experience.
Suppose you look into the results and notice that a big chunk of the negative responses is under the Billing tag… then something must be wrong with the billing process. Now, with this new insight, you have the power to reduce the number of negative responses.
You know the value of surveys in today’s market, but extracting it can be a challenge (both financially and physically) if you have to do it manually. You’ve been putting off that mountain of survey responses from SurveyMonkey, trying to figure out a way to get through them as quickly as possible. Well, now you know howWith text analysis you can turn that mountain into a molehill in just a few minutes.
With text analysis, your team can reduce the time spent on analyzing survey responses and focus their time on making decisions and taking action.
The answers to empower your team are at your fingertips. Still not convinced? You can request a demo and our team will help you get started with text analysis for SurveyMonkey.
May 27th, 2019