Previously we discussed the importance of open-ended data in gaining a true understanding of your customers. But how do you efficiently draw actionable insights from such a huge pool of data? To obtain information from open-ended data that is actually useful to your business’ strategy requires thorough analysis.
To do this manually, data analysts can spend days or weeks poring through 1,000s of responses to identify key issues and themes. Not only is this highly labour intensive, it can also be highly subjective, with different analysts interpreting feedback differently and prioritising different issues.
This is where the power of Natural Language Processing (NLP) text analytics comes in. Text analytics is a powerful tool to gain actionable insights into your customers and a strong investment in improving your customer experience.
It's also important to note here that, despite the way it sounds, it's really not that intimidating or difficult to use!
What is Natural Language Processing?
NLP is a longstanding field of AI research with a complex history, generally thought to have started in the 1950s with the Turing test. It seeks to create an understanding of human language as it is used by people in everyday life. Since the Turing Test, new technologies have emerged that aim to facilitate a deeper and more comprehensive understanding of text data.
Today, the two most-common approaches to NLP are rule-based (also called human-driven or supervised) systems, and statistical (also called machine learning or unsupervised) systems. Both systems seek to create meaning out of the data and are done either manually (hand-coding or machine-coding) in the case of supervised systems, or via machine learning in the case of unsupervised systems.
Gaining Insights with Text Analytics
Supervised (manual coding) text analytics requires the analyst to manually sort text data into a set of categories they've decided they want to track. For instance, for an online women’s clothing retailer, categories might include style, fit and shipping. The process is extremely laborious and time consuming.
Using computer-assisted manual coding to help apply categories to the data can speed up the process, but you still need to handcraft a ‘codebook’ of categories that you are interested in tracking and, most importantly, you are responsible for keeping that codebook up-to-date! Whether the coding process is entirely manual or computer-assisted, the result is the same: you get frequency and co-occurrence statistics of codes which you can examine with the goal of identifying insights. The problem with a rules-based approach like this is that it comes with inherent challenges and limitations, including human biases, time to results, changing language in the text data and complexity of the process.
Enter the unsupervised approach.
Recently, new technologies have emerged in the unsupervised machine learning domain that aim to facilitate deeper and more comprehensive understanding of text data with a fraction of the human effort.
Key amongst these is ‘topic modelling’ or ‘language modelling’, which was designed to make text data more accessible and more understandable as technology has advanced. It's an alternative to categorization, a way to find and trace clusters of words (called ‘topics’ in shorthand) in large bodies of texts rather than specifically looking for pre-defined categories.
Using topic modelling technologies, a language model (what might be considered a codebook in manual coding terms) can be generated based solely on the data uploaded for analysis. All conventional statistics such as frequency and correlation are automatically calculated for you.
Depending on the system you decide on, little to no setup is required and it will begin to pick up recurring themes in your data without guidance from the analyst in minutes rather than days or weeks. This method also allows for a much higher degree of control over the language model that will then evolve over time with the business.
One of the main benefits of emerging unsupervised text analytics software is the control it affords analysts in how they report insights out to the rest of the business. A quality unsupervised text analytics system will show the analyst all they concepts / topics / themes that are contained in their data. It's then up to the analyst to track over time for reporting purposes. Advanced systems will even allow the analyst to customize what the machine has found to match their domain knowledge and report on this instead.
In a retail example, this might be joining size and fit into one topic called sizing. If the issues customers are talking about changes over time, the system will automatically detect and report these changes to you, without any manual intervention from the analyst. This is a much better outcome than a manually maintained codebook of categories.
Combining these features ensures organization's can track issues that are most important to their business, without missing out on important new issues as issues emerge for their customers.
Getting the Most Out of Your Data
Unsupervised text analytics are the key to properly utilising your open-ended data, simplifying your analysis process and reducing the manual labour required to analyze your open-ended data. Unlike manual analysis, the process is repeatable and objective.
Different people feeding the same data into the system multiple times will all produce the same results. Not only does this provide a solid grounding for decision making on where to best focus your resources to improve customer experience, it also allows you to monitor changes in this data over time.
The ability to track open-ended customer feedback over time is where businesses can realise serious long term gains. By analyzing the trends in topics, both positive and negative, you begin to build a picture of how your organization’s customer experience is performing. It provides visibility into how internal CX projects are affecting the customer’s experience, and can be used to correlate this with other key business metrics. With this link created, it is much more likely that future CX projects will succeed as they have been developed based on hard data rather than gut feel.
Advances in text analytics software are allowing increasing numbers of businesses to realise the potential of their free text data from customer experience survey feedback. If you are collecting hundreds or thousands of NPS, CSAT or other customer feedback surveys each month and manually coding the responses, it’s time to invest in your CX strategy and upgrade to text analytics software.