At the Utility Analytics Institute, we're always listening. And one thing we've heard from the industry is a need to better share the stories of what utility companies are doing with analytics. In response to this need, we're launching a utility analytics project database that will be a part of our growing Utility Analytics Institute community. Interested in adding your analytics project to the database? If you have a great story about the analytics work you've done at your utility -- or know a client who has done some great work -- we'd love to hear from you. Interested in accessing these stories? Just let us know, and we'll be happy to get you more information about the database. Here's a snippet of some of the work we're doing for the database. This project profile focuses in on customer engagement analytics, but it is important to note that the database covers a wide variety of analytics projects.
Background and introduction
A large western electric utility has long been utilizing advanced analytics, and more specifically, predictive modeling, to identify and target the best prospects for value-added services and programs. These services and programs include: levelized billing, renewable power, electronic billing, preferred due dates and automatic payments.
About 10 years ago, the company began an effort to improve customer targeting by developing a marketing database and applying advanced analytics. Effective and efficient customer engagement sets the utility apart.
Prior to using predictive modeling, the utility used ad hoc profiling analysis for customer targeting. Customer groups with different characteristics, such as electronic billing customers and non-electronic billing customers, were compared. Variables that seemed to indicate good targets for a particular program or service were used to identify customers as targets for marketing campaigns.
Although the utility still relies on data queries and human intuition, the company uses advanced statistical modeling and tools to more precisely determine which combination of attributes are most predictive. The output of the statistical analysis is an equation that can be used to score all customers based on their relative propensity to enroll in a particular service or program. The process is repeatable and is used to target customers for a wide variety of programs and services by including only customers with high scores on promotional lists.
Behind the decision to use predictive modeling was a desire to improve the efficacy of marketing campaigns and to enroll customers in programs and services at a lower cost.
Customer satisfaction was an equally important driver because of concerns about sending unwanted direct mail and email to uninterested customers. Although email communications are not particularly costly, customer satisfaction is paramount. The utility identified an opportunity to leverage its own customer information and third-party data. The idea was to apply the same type of advanced analytics the financial services industry uses to improve marketing efficiency and to deliver a more satisfying customer experience. This initiative was also well aligned with the program and service-specific goals of the utility's product managers.
The data utilized for this analytics application resides in a database, a central repository similar to a customer-oriented data warehouse. The relational database was developed via an iterative process and has grown in response to internal client needs. The company implemented an archiving process in 2005. Today, the company enjoys a very flexible 350-gigabyte customer database -- not including interval data. The database runs on a single server, and advanced software runs logistic regression routines and builds the predictive models.
The marketing database is not only the primary data source for building customer targeting models, but is also a repository that supports other applications. For example, it provides internal operational support for marketing programs and is used to develop direct mail lists for meeting day-to-day customer communications needs.
The database includes the utility's own data, including customer information system data, web data, outage data, data from certain distribution systems, and substation data. Third-party data is also appended. For example, demographic information from an outside provider is appended to customer records twice each year. Demographics include how many people live in a household, their ages, gender, occupations, income and other attributes. The data provides about 85% coverage and significantly improves the utility's ability to understand and model customer behavior.
At this utility, an effort is under way to examine how trended data may further improve model performance. Previously, models developed internally were based on a snapshot of data at a point in time. An outside provider recently partnered with the utility to build four models as part of a pilot. The provider recommended developing the models based on 24 months of data, explaining that a customer's propensity to enroll in a program or to let bills become past due changes with time. Because the archiving process was implemented in 2005, the historical data needed was available. The hope is that trended data will result in better models, though a different level of sophistication than currently exists within the utility will be needed to develop these models.
The technology lead and operational owner of the database is the manager of customer research and analytics. The utility may favor an enterprise-wide application at some future time. If so, the company would establish a cross-functional team that includes the IT organization and other stakeholders.
Prior to implementing predictive modeling, the utility used more elementary methods of targeting. Queries were built to identify users versus nonusers of a particular program or service, purchasers of other programs, or customer payment history. Variables then were selected for targeting purposes. This process has changed substantially as the utility adopted a more sophisticated way of targeting customers. Logistic regression modeling and advanced tools are used to develop more precise, statistically sound customer targeting and to achieve "lift" compared with previous targeting methods.
Members of the customer research and analytics team foresee the potential for significant business process effects from applying advanced analytics. For example, the manager of customer research and analytics envisions that an analytics-based recommendation engine could one day be used to change the nature of customer interactions by prompting customer service representatives (CSR) to offer a program or service -- renewable power, a levelized billing plan or a payment extension plan and so forth -- designed to appeal to an individual customer's specific needs.
Applying advanced analytics is facilitating conversations and collaboration among many groups within the utility. For example, the earlier mentioned pilot project and the potential for better targeting has drawn attention from employees throughout the company. A recent session conducted to review the pilot project's results and potential applications saw participation by employees who specialize in corporate communications, product development, product management, channel management and market management.
Even with these examples of collaboration, the reality is that it will take time and persistence to build awareness of ways advanced analytics can deliver substantial operational and customer experience improvements. It's an evolution that requires top-down support and necessitates cultural changes in the way utility employees approach their work.
Customer satisfaction (i.e., finding ways to enhance the customer experience) is a strategic priority for the utility. This objective has been a primary motivation for developing the marketing database and starting the predictive model-based targeting initiative.
More sophisticated statistical customer targeting models support corporate goals to accelerate the adoption of certain programs and services, such as customer self-service and renewable energy programs that rely on clean energy such as wind power.
The company's overall customer strategy executive sponsors this initiative. The customer research and analytics team is part of the customer strategies and business development group. The utility expects this group to deliver an overall first-quartile customer satisfaction rating (CSAT). Even though the utility doesn't measure customer satisfaction related to marketing campaigns individually, it believes that more effective targeting will make positive contributions to this objective.
In addition, the company set a goal for each marketing campaign to realize "lift" as indicated by enrollments and improved campaign efficiency measured against historical results. The utility has an objective to lower overall marketing spending and cost per enrollment. Although objectives are generally qualitative at this point, the utility intends to make them more metrics-driven.
Improved targeting supports program-specific goals, such as increasing electronic enrollments, increasing renewable energy enrollments, and decreasing past-due bills. For example, the utility used predictive modeling for a 2011 electronic billing campaign. The billing and payment programs manager's feedback was as follows:
"Using a paperless billing customer profiling model and developing emails targeted at customers with a strong propensity toward paperless billing caused enrollment numbers to spike."
The utility plans to use predictive model-based targeting for any enrollment-based service from now on. Future possibilities include: a critical peak pricing program, outage alerts, and high bill alerts when bills are too high.
Two employees on the customer research and analytics team built the first predictive model and several subsequent models. Because of human resource constraints, a consultant hired a couple of years ago developed a suite of models for five products and services.
Only one employee serves as a dedicated resource. A capable database developer/administrator is dedicated to the marketing database operation and development. He manages the system, provides a wide variety of analysis services, and assists others in writing queries. Other employees are involved in model development, although they are not dedicated to model development, and they have other responsibilities.
The utility has not yet put formal change management procedures in place. Eventually, such procedures will be needed. The utility envisions that analysts and non-analysts alike will have better tools and will be able to access and take advantage of advanced analytics. Changes to the core system are under way, and they take precedence over analytics change management procedures. Some of the new core systems must be in place before effective change management procedures can be implemented because they depend on systems from which the data is pulled. It will take time and resources to build the processes and tools required to facilitate data management and data governance.
So that's a taste of our forthcoming utility analytics project database. If you'd like to learn more about the database and how to get involved, please don't hesitate to reach out to me at email@example.com.
Thanks for reading!
H. Christine Richards is the director of knowledge services for the Utility Analytics Institute, a division of Energy Central. You may reach her at firstname.lastname@example.org.