Coverage of this session by Evan Perkins of SocialMedia.org. Connect with him by following him on Twitter.

12:20 — SocialMedia.org’s Kurt Vanderah introduces Xerox’s Tong Sun.

12:22 — Tong: During the past several years I’ve worked with social groups at Xerox. There are many challenges with social media for business. There are lots of data available but very little actionable insights available. This is because it’s very difficult to measure! There’s a big gap with measures and connecting to business goals.

12:23 — Tong: During one example, we had to transform a traditional call center into customer experience management. The call volume fell flat and we knew we needed to change. We had challenges with growing opportunities, improving customer satisfaction, automating processes, and business modeling. They discovered that cost per call was $0.35, with customer sat. dropping and processes not in place.

12:25 — Tong: When we started we first identified the strategic context: What are the customer requests? Desired features? What do our competitors offer? What are customers saying? At that point, it’s all about the brand. The call center is an engagement touch point so we need to leverage the resource to do more than just resolve issues. We discovered that much of this data was unknown.

12:27 — Tong: From there, we looked at social media as a data source to help us discover these unknowns. For example, in telecom we listened to cell signal strength in specific areas and were able to alert location-specific business units to help them be informed and provided them with a response strategy.

12:28 — We must keep people in the loop!

12:29 — Tong: When then moved on to discover “how,” what’s the efficiency of the business operations? We outlined the process to discover how to disrupt and improve in order to optimize. We engaged social for process context and optimization. We brought in a computational linguistics expert to help us with machine learning — the purpose was to find sarcasm! Very difficult to measure with just a key word.

12:31 — Tong: We then setup a skill-based automatic workflow. We would begin to route calls to specific team members with expertise in the needed area.

12:33 — Tong: Listening was based on machine learning approaches. Our listening metrics involved granularity, depth, accuracy and coverage, and signal/noise ratio. This helped with early detection & prediction to add into our call centers.

12:35 — Tong: At the early engagement stages we recalculated metrics for social. We used the domain expert to label some of the engagement on specific threads.

12:36 — Tong: The call center was the transaction-based pricing model, but with social we had to change to a value-based model looking at things like engagement, sentiment conversion rate, crises prevention rate, timeliness of engagement and speed.

12:37 — The good old rules apply: Bad news spreads faster than good news! Negative word of mouth impacts more than positive word of mouth.

12:38 — Tong shares some key takeaways:

  • Don’t treat social media in a silo
  • Focus on customer problems
  • Link your measurements with actions
  • Be adaptable and agile (data-driven with people in the loop) and willing to be constantly changing

Q & A

Q: How do you do this with brands where there are not clear solutions?

A: Tong: You have to live and breath with your people. Do quick pilots and setup quick listening programs to learn about them and their challenges quickly. Build a relationship and you build out a process for communicating.

Q: Regarding volume, how do you identify larger customer issues? How to take it from individual interactions to larger scale?

A: Tong: People have to inter loop. There isn’t a magic bullet here, so you need to aggregate your big customer issues. First understand who your important customers are (because of all the noise). You learn this gradually by listening and going into “discovery mode.” Leverage your domain expert and SMEs.

Q: I am intrigued by your unified identify. How did you work with your call center people to create unified entities of people or issues?

A: Tong: This is difficult. It’s all about inferencing. With social you’ve got to look at the individual against the population. Infer their interests and location  by who they followed, what they tweet, etc. Then look at demographic in your call center. Then correlate that user and know they will probably act/respond in similar ways when compared to population.

Q: What tools do you use?

A: Tong: We use a research prototype. Not a vendor tool, though we do have Xerox Business to commercialize it.

Q: Do you use listening tools married with text analytics tools?

A: Tong: Yes, we combine listening and analytics. MPATH is the tool we use. This isn’t a turn-key solution.

Q: Any traps we should all watch out for?

A: Tong: You need to challenge the way people measure things if you don’t understand (oftentimes, they don’t understand either!). Redefine what is necessary and do not push forward until all measurements are understood.

Q: Are you using tech to automatically cluster the issues you’re seeing?

A: Tong: Yes, clusters you can always do automatically by words, occurrence etc. Classification is a little bit more difficult because in different product domains sentiment differs. Very hard to do this automatically; you need domain experts. What we do is pick a small sampling from a small amount of the data and have a domain expert look and analyze the data.


Get our free weekly newsletter

A short email packed with updates on what big brands are doing in social media.

Never display this again