Benefits of Multi-Tiered Clustering
By Edwin Margulies Clustering, or categorizing "like issues," is an effective way to accelerate the outreach and resolution of customer care issues over social channels. By categorizing like issues, your social care team can more easily identify hotspots. With built-in Natural Language Processing, and a rules-based platform, you can automate routing and next best actions related to these clusters, too. But before you lump like issues into a single dimension, consider the benefits of multi-tiered clustering. First, What's a cluster? A cluster is a group of words and phrases that make up a sub-topic of common ideas. In the context of social engagement for customer care, these sub-topics represent the undercurrent of a general conversation shared by many people who are posting socially. Automatic clustering of like topics makes it a lot easier for your social care team to do outreach. Instead of having to read the entire content of each post, you can just eyeball the cluster name. Across your whole social care team, this can save many hours each day. Initial Static Cluster Analysis When you first set up a social engagement for customer care system, one of the important steps is to analyze the "word clouds" representing the clusters that are gathered by the NLP/NLU (Natural Language Processing) Engine. Modern systems allow you to run the NLP engine against all of the incoming social posts so the engine can start to categorize clusters of like phrases. This is an important step because you can use this initial data set to set up your "First Tier" of clusters. The idea is to take a look at the word clouds that have been gathered and group them together into issues that are persistent for your business. For example, if you are running a service business, you may observe word clouds falling into categories such as "billing," "service plans," "usage," "sales," "complaints," and "technical support." For the most part these issues are static, persistent issues that your business deals with every day. They are not really trends, but common categories you have probably already mobilized a customer care team around. By way of example, the phrases: "blew out my usage," "my bill is high this month," and "may need to upgrade my usage plan" all share a common theme of usage. Seeing these phrases in the first truanch of word clouds will help you to entitle a static cluster called "usage." You can validate what the persistent issues are if you look at the word clouds over days and weeks and they are still the same. Clusters that show up outside of these persistent issues are for the most part "trending" issues and they can be used to establish a sub-text or theme UNDERNEATH the static clusters. Trending Issues - The Second Tier The reason why having a second tier of clusters is useful is because you can trigger rules to take action on those trending issues. In addition, you can do this without disturbing your SLA reporting and analytics, which are based on more persistent topics. For example, you will surely want to report on how many social posts fell into the persistent categories of "billing" or "complaints." But underneath the static cluster of "complaints" there may be certain types of complaints that you don't want your social care team working on. Let's say, for example, you don't want them working on complaints that are not actionable like "I hate the color of XYZ Company's logo." With two-tiered clustering, you could still register the number in the static cluster "complaints" but peel-off and disposition as "closed" those complaints (like the color of your logo) as not actionable. Or you could even create rules to trigger those complaints as "transferred" to a brand manager or someone else to whom the complaint is actionable. Other Tier Two Cluster Triggers Imagine being able to identify a trending topic and then filter it based on sentiment, influence or how long it's been waiting on a response. That's the power of combining a NLP engine with a business rules engine. Here are some use cases that will help you to think of more for your business: - Trending issue: New store opening and being under stocked on a certain item. Filter on angry sentiment, and route influential posters into a special work queue. Outreach with a "rain date" coupon or other offer. - Trending issue: Competitive New Product Announcement. Filter on happy sentiment, and key words such as "buy" and route to a special "Sales" queue. Outreach with a competitive offering. - Trending issue: Spate of Product Failures with Common Theme. Filter on all sentiment, and key words and phrases: "just bought," "broken," "Help," etc. Disposition all from this word cloud into a special service queue. Outreach with an "RMA Hotline" URL. Conclusion Two-Tiered clustering is an effective way to approach the categorization, and automatic filtering and disposition of social posts for your social engagement for customer care operation. Take the time to properly characterize persistent, static business issues as your first tier of clusters, so you can take special actions on the trending topics happening around these common business themes. This approach will save your social care team many hours each week and improve customer loyalty at the same time.