This article originally appeared in the Nov. 2012 issue of CUSTOMER
Big data, put simply, refers to datasets that are so large and diverse that traditional data management and analytics tools are incapable of handling them. Big data can be either structured or unstructured, which is important because only about 10 percent of all enterprise data is structured and formatted to cleanly fit into databases and logical organization schema.
The vast majority of enterprise data is unstructured: e-mail, Facebook (News - Alert) posts, tweets, chat transcripts, call center interactions, website activity, support forum conversations. All of this, gathered from multiple channels such as your website, mobile applications, contact centers, e-mails, and social media interactions, form more than 90 percent of the data deluge that businesses routinely collect. Arguably, this 90 percent contains the more interesting and valuable data about customers, since these conversations are a direct expression of customer likes and dislikes. This vast majority of customer data goes untapped, un-mined and unseen.
This is increasingly important, because consumer behavior has changed too. Customers want intelligent interactions with companies, and they expect these experiences to be seamless and consistent across channels. The key to delivering these experiences lies in the data. To deal with data at this scale, certain emergent technologies are gaining prevalence. These include distributed databases and file systems, and programming tools such as Hadoop that allow easy manipulation of data at scale. Other sets of technologies such as machine learning, predictive modeling, pattern recognition, etc., leverage big data tools to enable us to identify trends and learn at scale.
When put together, all these new tools and technologies allow us to use big data to transform customer service, engineering a quantum leap from the old, reactive way of doing things, to new, differentiated models of customer experience.
The Rise of Consumer Experience
Big data is already a part of your consumers’ everyday life. Shaped by technology, customers today expect experiences (enabled by data), that simplify their tasks and reduce effort. This expectation is belied for them almost each time they reach out to a company to seek service or support. Every voice, chat or web interaction still starts with a clueless agent or an opaque website that seems to ask the customer: Who are you, and why do you want to talk to us? Seeking customer service from the enterprise continues to be an inherently painful exercise for most customers.
There is this big gap between the actual and potential quality of each and every customer interaction. Businesses have data. Deploying it to gain insights into the consumer and applying these insights back to every interaction – now there’s the rub.
While many enterprises are becoming increasingly adept at using analytical tools and big data to increase transactional sales, customer service and support interactions have not received an equal share of attention. Moreover, customer data is yet to make the leap from a tool to enhance some transactions, to a real strategic asset.
Your customer data can tell you who your customers are, their history with you, their past interaction data, how their current journey on your website or at your toll-free number is going and if your customer is vocal about you on social media.
Aggregated together, with the right tools and methods this data across thousands and millions of customers reveals patterns, trends and insights.
A Framework for Big Data
Identifying the available types of data is one thing. Applying it in real-time to drive outcomes is another. Companies must apply a simple yet powerful framework to describe the philosophy behind this application of data analytics to customer interactions.
Determine who you are talking to, what they are trying to get done, and when they require help. Consumers reveal their intent through their behavior on the channel. For example, a journey on an electronic retailer’s web page reveals information about the products that the visitor is interested in purchasing.
Blending current journey data with the identity of the individual and his/her context from recent interaction history across channels can substantially drive up the intent prediction accuracy. Factoring location data into this can further boost the prediction, and we can compare behavior in a single interaction with the behavior of thousands of similar customers using various statistical analysis and machine learning techniques. We can also identify the exact points in the journey when the customer is most likely to take an action or need an intervention. All this can happen in real time during the regular course of the customer’s journey without requiring them to provide any explicit input.
Once you have determined a customer’s identity and predicted his or her intent, the data can help you decide the best interaction type to make the experience simple and fruitful for your customer. This includes decisions on the right channel (or channel combinations) in which to engage the customer. One of the powerful tools that modern technology places at our disposal is the power to conduct multiple experiments with samples of our customer base to test responses. These tests, which follow a design-of-experiments framework, are known as multivariate tests. They allow us to identify the best presentation methods and channel combinations to engage with consumers and resolve their problems. For example, today in the web world it is very easy to segment visitors and treat each segment differently. Different web page designs can be shown to different people to quickly understand customer responses to different designs. We can leverage such techniques to determine the experiences that drive the most effective customer experiences, leading to reduced customer effort and successful resolutions.
The most critical step in this framework is applying the results from the previous steps to making subsequent interactions better. The critical fuel that makes subsequent interactions better is more of the data itself. Smart customer service applications can use the data that they generate to self-correct, automatically learning from each interaction to improve customer targeting, prediction accuracy, and outcomes.
Current predictive models used in customer service are rudimentary: the most sophisticated typically consist of basic business rules that segment customers using broad brushes. For example, people who spend more than 30 seconds on a particular page may be exposed to a particular offer. In contrast, our machine learning models are trained on interaction level data that spans customer touch points from web click-stream data to detailed IVR interaction logs, to transcripts of chat and voice interactions with customer service agents. This is in addition to the transactional and CRM data. These models are evaluated dynamically, in real-time, based on the customer's historical activity as well as the current interaction with the mobile device to provide predictions about the customer's current intent and provide context to the experience engine. As a result, these models are completely customized to that particular customers’ situation. With such an approach, we can better understand the preferences, interests and needs of consumers to predict the reasons for their interactions with a company.
Due to the difficulty in processing unstructured data, the standard industry practice is to analyze customer surveys as a proxy for interaction data. In contrast, unifying platforms must leverage a big data infrastructure that mines 100 percent of customer interactions, including customer chats and phone calls. This then allows for learning at scale from all customer interactions, providing richer data for the predictive models and, in turn, increases their accuracy. It also allows for analysis of agent behavior in great detail and scores every single interaction on multiple dimensions such as agent soft skills, agent performance and other drivers of successful outcomes. In turn, this provides powerful insights for contact centers and can be used to drive decisions on training, staffing and agent management.
The fusion of unstructured data from interactions with customer service representatives with the other elements of the customer journey also captures any intent prediction errors, thus creating a self-correcting feedback loop. An actual conversation of a customer with an agent is probably the best data to identify true intent. The insights gained from mining all service representative interactions are also used to improve the resolution experience of self-service applications.
Enterprises need to reorganize around the relevant sets of technologies and processes that lead to higher customer centricity and superior, differentiated customer experience delivery. At the same time, the required velocity of change, combined with the relatively slow response times of typical IT cycles, demand alternative delivery mechanisms that can accelerate your transition without requiring significant capital investment.
Ravi Vijayaraghavan leads the analytics and data sciences organization at 7 (http://247-inc.com).
Edited by Brooke Neuman