Communication service providers aim to enhance customer experience and build strong and long-lasting relationships with their customers. This has become increasingly difficult as customer interactions now occur across many channels. Hence, it’s important to understand customer behavior across all channels to create the best experience for each individual. Join us on August 5 for a webinar with Hortonworks and Apigee to learn more.
In today’s guest blog post, Sanjay Kumar, General Manager, Telecommunications at Hortonworks, and Sanjeev Srivastav, Vice President, Data Strategy at Apigee, discuss how service providers can capture and visualize customer behavior as a graph connecting the interaction points such as IVR, chat and call events, and combine it with network data to predict future call or chat patterns. For such analysis, many telcos have invested in big data infrastructure such as the HDP platform, and some have already begun to harness the myriad of data gathered from multiple internal systems in data lakes. Furthermore, there is the realization of the importance of assembling a single, 360-degree view of the customer, with customer-focused data sets.
Telco companies’ next challenge is to answer questions such as:
Means of answering these questions must be grounded in the reality of changing customer behavior and operational metrics. Examples presented in this blog used the HDP platform coupled with the Apigee Insights software to not only provide reliable predictions, but also tools useable by the business managers. This allows the businessperson to understand how customers in a similar context have engaged with the company, and how effective specific interventions have been.
When telco subscribers visit a support site or use a care app, call customer support, experience an outage, or express an opinion on a social network, they are taking steps on a “customer journey,” which represents a sequence of events experienced by a group of customers; in other words, it represents various paths in the overall customer behavior graph.
The traditional approach towards understanding overall customer behavior involves:
These approaches are limited in their ability to answer questions like these:
The ability to ask questions like these (which involves walking forward and backward in time, across events from multiple channels of data) and to build predictive models based on these customer journeys provides a foundational layer for the analytics capability.
Once the journey analysis and predictive models are in place, the business and IT teams need to deploy the solution at scale. In this context, Hortonworks and Apigee have partnered to bring to market the Apigee Insights technology, a unique method of building aggregate behavior graphs on the Hadoop platform. Here we’ll describe work done for a communications provider, using data including customer call center records, operations data, and the network quality data.
The picture above shows a schematic graph that represents customer behavior. Note the sequential placement of events in the behavior graph. Such representations are used to compute the likelihood that a specific customer would be a repeat caller in, say, a 21-day window. Furthermore, based on the journeys experienced by a significantly large population of other customers, the system also provides scores for the likely reason behind that next call.
The specific business unit in question here receives about 100,000 calls a day, from a customer base of approximately 10 million. Over time, one can imagine the almost infinitely large number of paths that can be generated to represent the customer engagement. By applying certain threshold parameters that govern the amount of activity represented by the various possible paths, one obtains a set of “interesting” paths—but even those can number in the millions. It takes the power of big data platforms like HDP 2.x to manage the storage and computational power needed to handle this variety and volume of data.
In the screenshot below, you can see some examples of journeys that end at an “escalation” event, experienced by about 226,000 customers. Clues from these journeys and details of the customers who experienced these journeys are directly useful in effecting change, say, by increased training of support representatives or by proactive outreach to customers.
Furthermore, one can also inquire from the system the exhaustive list of paths that customers potentially took between two events of interest, and noting the relative fractions of customers who follow the various journeys. Such a bird’s-eye view provides an understanding of customer behavior that is very hard to glean simply from a set of customer attribute values.
Besides the need to visualize past customer behavior, it is critical that predictive models keep pace with changing trends. Models that use a combination of customer profile elements and event activity data are better suited for the twin tasks of predicting the likelihood of repeat calls and the likely reason of the next support request.
Apigee’s behavior-based predictive models, which don’t require the conversion of event data into customer profile data, and also preserve the information inherent in the event sequences, provide excellent predictive power.
Change is inevitable, and data-driven insights help companies better manage change. For example, data about the intensity of customer activity on specific paths quantifies the impact of changes to that particular path or process.
By knowing a customer’s recent journey segment (she experienced a service outage, has called into a support center, and her service just got restored), the company can delight her by sending an automated outbound SMS.
Access to customer journeys at scale provides business analysts with a feel for the problems and opportunities for improving the customer experience. Additionally, if the predictive modeling is done using data structures that represent the journeys, analysts also get a sense of customer behaviors that most influence the predictive modeling; this is hard to achieve from a black-box predictive model.
Lastly, once predictive model have been developed, the overarching goal of improving customer experience can be reduced to specific actions that are appropriate for specific groups of customers. Call center agents or supervisors can execute these actions via existing or new applications; an intelligent API platform can further streamline the “last mile” problem of delivering predictive model scores to the applications that need them.
Without needing “features” based on insights shared by business owners, the model has high precision for identifying repeat callers (96% correct for the top 1% of the list of customers, sorted by their propensity of repeat calls); for each of the callers, it also predicted the top few reasons likely associated with the next call (actual reason of the call was ranked the number one reason for 61% of the top 1% callers).
As digital interactions dominate the enterprise landscape, data about how customers engage with companies is being stored in Hadoop data lakes, for it provides critical business insights. Apigee Insights provides a powerful method of representing such engagements as behavior graphs while taking into account the raw data about the time and sequence of activities. This representation is used for descriptive analytics (examining historical data) and predictive analytics. The last mile of leveraging predictive analytics to influence or enable customer engagement is facilitated using Apigee Edge. Customer care decisions can thus be much more targeted and likely to have a positive impact.