Southern California Edison | The 2019 Excellence in Marketing and Communications Awards

Describe the Challenge

In 2014, SCE resources delivered 6,783 responses to customers via Facebook and Twitter, and over the course of the next three years the total number of responses grew by 391 percent (to 33,350 in 2017). That hyper accelerated growth rate that was especially impactful on resourcing in the last of those years, as a couple of changes to our social customer service environment—notably the rise of Facebook messenger—contributed to the total number of responses to customers nearly doubling from 18,334 to 33,350 in 2017.

We had grown our dedicated social customer service team from one to two over this time period and, continuing at thisrate of growth, we'd project the need for additional resourcing needs possibly as early as 2019. This, problematically, flew right in the face of our company strategy that was becoming increasingly focused on resource efficiency.

So, what do we do?

Define the Target Audience

The target audience is the increasingly large portion of the 15 million customers of Southern California Edison who prefer to communicate with us via their smartphone (or computer) about issues such as outages or billing questions.

Describe the Solution

The solution was to implement a means of resolving customer service issues on Facebook and Twitter at the same high “concierge” level we had historically achieved without requiring additional human agents.

Solution: Sprout Social Chatbot

Goals

1. Maintain or increase our level of social customer service without having to increase staffing.
2. Allow chatbots to resolve simple inquiries so human specialists can work on more complex inquiries.
3. Provide 24/7 customer service on social to customers (even though we're only resourced to staff from 9am-9pm Monday to Friday and 10am-7pm Saturday).
4. When there is a crisis or a significant incident that generates many times the number of inquiries that we can respond to (with human agents), have the capability to get critical information out to all of these customers as quickly as possible, if not immediately.

Objectives

1. To keep our FTEs dedicated to social at two by keeping flat, or decreasing, the number of responses human agents provided via Facebook and Twitter.
2. To remain at or above our 2017 social customer service after-care survey overall rating by ensuring that customer needs are met (regardless of whether a human or bot is responding).
3. To be able to respond immediately to customers “after hours” with either an event-specific response, a resolution (if simple), a phone number for immediate assistance, or an invitation to provide a detailed inquiry understanding it would be resolved by a human agent until the following—or Monday—morning.

Describe How the Solution Was Deployed

In late 2016, the company was exploring chatbot solutions for its website, and in the early stages we considered merging social into their pilot solution, but there were technical as well as practical complications. On the latter point,the web chatbot was AI/learning based, and we believed that the level of service had to be higher in social media environment where customers have the ability to give and share publicly good—or bad—experiences right away.

At the same time, Sprout Social, which we'd used as our social community management platform for many years, was launching a bot that was “decision-tree” based, which we'd preferred for social media. We believed that having astructured, decision-tree based chatbot—vs. an unstructured/learning-based chatbot—would provide for a much more optimized experience given what we already knew about our social customers, which was that 77 percent of the inquiries from our customers were about only five issues: outages; billing; credit; programs and services; and our website.

So, we onboarded the bot in late 2017 and built out the decision trees and were fully operational as we entered 2018. As the year progressed, we created multiple versions of the bot and we activate those different versions as needed (examples: crisis bot, website down bot, holiday/we're closed bot) to allow us to communicate more immediately—and efficiently—with customers when there is a single issue that generates more traffic than our human agents can handle.

Describe the Qualitative or Quantitative Measures of Success

1.  We maintained our staffing at two and don't have a projected need to increase staffing at any point moving forward. While our total responses increased from 33,350 in 2017 to 41,812 in 2018, the total number of responses from human agents actually decreased from 33,350 to 24,146. The chatbot was responsible for 17,666 responses to customers, and just over a quarter of those responses (4,593) resulted in full resolution of the customer issue. The remaining responses required a human agent to step in, but only after the bot would have (in most cases) already gathered the basic information needed to research the inquiry (thus minimizing the time required by the human agent to resolve the issue).
2.  2017 rating = 7.8/10; 2016 rating = 7.8/10.
3.  The bot provides an immediate response with options to be redirected or to enter an inquiry.
4.  When we have a major outage (ex. 50K-100K customers without power) we can create a note specific to the outage and the bot will autoreply to every inquiry, allowing us to reach 100 percent of the customers with critical and specific information vs. only a fraction of that with human agents only.

List the Tools Involved

Sprout Social Bot