Mining Structured & Unstructured Data to Strengthen Customer Experience
How pairing ALL data sources for text analytics can reveal WHERE to focus, WHY to pay attention, and HOW to address what customers want and need.
By Jay Yeo, Data Analyst, ORI
In a recent blog post, we explored the immense reach, volume, and potential of the insights hiding in unstructured data—the type not easily categorized and neatly organized, such as data amassed through social media, community forums, emails, chats, surveys, calls, audio/visual files, open-ended responses, and notes fields. Though most organizations already have unstructured data at their fingertips, many struggle with how to organize and analyze these valuable data points to uncover actionable insights that can improve member or customer experience (CX) and boost revenue—particularly when combining unstructured data with structured data.
The value-add of unstructured data mining is quickly gaining traction, and it isn’t limited to any particular industry. According to an Ernst & Young study of asset managers, nearly half of those surveyed in 2016 were not using alternative data (i.e., unstructured data)—but by the following year, 78% were already using or expected to use nontraditional data. Although unstructured data alone holds a plethora of customer and member insights, key to utilizing this invaluable resource is combining it with structured data (the type stored in Excel files, CRM or AMS platforms, and event registrations). While structured data can identify the “what”—trends, KPIs, and changes in the bottom line—unstructured data, when paired with structured data, provides insight into the “why”—the reasons for change and subsequent action steps to be taken to address issues and generate bottom-line impact.
By understanding the “why” behind the “what” through mining all types of available data, organizations are able to gain a more granular view of not only where they need to focus attention but also why they need to address it and, therefore, how they can address it. For example, suppose that the structured data of a food supply company’s automated end-of-checkout survey demonstrates a falling satisfaction score. More specifically, it shows a falling satisfaction score among those customers who place bulk orders. Without unstructured data, it is easy to spot the falling metric but very difficult to pinpoint precisely what the problem is, let alone what to do about it. Incorporating unstructured data, on the other hand, might reveal that the level of effort (LOE) expressed by respondents discussing the online ordering process is very high and that respondents are complaining about the inability to add multiple orders of an item to their cart at once. Instead, they are having to navigate away from the automatically linked cart page back to the individual item page to add more of the same item. In this sample case, combining unstructured data with structured data allows the company to pinpoint precisely where and what the problem is to better address the issue. In this case, the company could, by making a small adjustment to its online ordering website, reverse a negative trend to see bottom-line improvement.
While it may initially seem daunting to have to put two different data sources together, pairing unstructured with structured data can be accomplished if you break the task down into steps. In fact, many unstructured data sources already contain structured data:
- Open-Ended Response and Notes Fields: Most open-ended response fields are already combined with all of the structured data points captured in the data source, whether a survey, a questionnaire, or a medical document.
- Call and Customer Service Centers: Customer service recordings and transcriptions typically come in datasets with associated structured data, such as customer ID, employee ID, time and date of interaction, number of transfers, amount paid, and items ordered.
- Online Communities, Email, and Chat: All of these forms of online interaction generally come with time stamps, usernames or email addresses, volume of messages, and possibly interaction type, customer profile data, and other associated information.
- Social Media: Even social media—the unstructured data source most often associated with the least amount of structured data—comes with structured data fields that include a combination of a timestamp, username or handle, age, location, education, volume of posts and interactions, number of followers or subscribers, number of likes or retweets, number of stars given, and more.
Tying structured and unstructured data together can be as straightforward as not separating them from one another prior to analysis and then making use of the full range of data provided by a data source. This assumes, however, access to well-developed unstructured data mining technology, such as the type provided by Clarabridge (now Qualtrics XM), which ORI has implemented for numerous clients to analyze open-ended customer feedback and drill into themes, LOE, sentiment, and emotion. Doing so not only ensures that an organization is using all possible data points but also adds depth, direction, and impact to the analysis by uncovering the “why” behind the “what.”