Clustering: How does it work?

Why is clustering important to understand? Read on to find out why clustering is a crucial step towards automation from our CEO Reetu Kainulainen.


Outline:

  • Definition of clustering

  • The main benefits of clustering

  • Applications and use cases for clustering

  • How clustering works at ultimate.ai in 3 easy steps

Introduction

As customer service teams around the world race towards automating their repetitive processes, there’s a lot of confusion out there. Here at ultimate.ai, we love the nuts and bolts of automation because they bring order to chaos -- and that’s just what clustering does. It’s such an important topic that we interviewed our CEO and Co-Founder, Reetu Kainulainen, about it. In the interview below, he breaks clustering down to its bones and gives a clear and thorough overview of why it’s a critical step for any customer service team preparing for automation.

Definition

What is clustering?

Clustering is a way to bring order into chaos. If you think about customer service and NLP, it means the chaos is basically all the historical conversations you have stored somewhere, but you might have millions of them and it's very hard to automatically try to understand what's going on. In many organizations, they spend time manually going through each conversation, its messages and then tagging them to understand what is being said.

Imagine you have a box of legos that you just pour on the floor. You have this huge chaos of different sizes and colors all spread around. With clustering, what we try to do is automatically cluster them into coherent groups. When it comes to legos, it might be that after clustering, you have the same colors in little nice groups or they might be the same sizes or whatever the pattern there is, so now we have order, and now it's easy to kind of see and analyze, ‘Oh, I have 20,000 yellow legos and 5,000 red legos.’ When it comes to customer service, we do the same thing for the customer service messages and conversations.

So after clustering, you can see what are the largest clusters, which means, what is the number one question your customers are asking? What is the second, and how many of those questions you can also cluster the agent messages -- what is the average answer that agents are giving, and this is a very crucial step.

If you want to even consider automation, you have to first know what to automate.

Main benefits of clustering

Why is clustering useful?

With clustering, you can have a really data-driven approach to understand what is going on in your support organization, or in the whole business itself, because now you understand what your customers are saying. One interesting thing we see with many companies is that they have, for example, FAQs on their website [for their customers] and FAQs for their support teams -- and there's usually a huge difference between what companies think their most frequently asked questions are, and what are the actual results.

When you really look at the data, you go through the past 10,000 conversations and we've seen cases where companies think that the FAQ is about 50-60% of the total volume, but in reality, those questions actually consist of 2 to 4 percent of the total volume. So imagine running the organization or the support team without really understanding what are the biggest processes or biggest questions you have to tackle.

Customer Service Automation: Use Case

If you're thinking of starting to automate some of the conversations or the cases in your customer service, it's really important that you know which ones to pick first. There's a lot of technology out there that builds customer service automation but the problem is, it's almost like a blank canvas: where do you start? What do you do? What is the first question you start to automate? Do you look at your FAQs? Maybe they're not that accurate, so with clustering, we can actually show you, here's the 100 most common questions you get, and maybe we show you that the number one question is 10% of the total volume, so now you really skip the step of kind of try to figure out what we should do, and you immediately go into, ‘OK, how do we want to automate this? Let's pick the ‘where's my order?’ question and then map out the process, then [ask ourselves] how can we automate so that the end-user experience is great?

This is really the power of clustering, so you can fast-track the discovery phase or figure out what to do -- it will show you just pick a few of the cases you want to start automating. It also ensures that the results are more certain so you know what to automate but also you know how your customers are asking the questions.

With clustering, you really skip the step of ‘what should I even automate?’ That’s because instead of trying to guess what customers might ask, you already have that data. With clustering, you can tap into those insights and really see where to start. After all, you don't want to start by automating 200 questions at a time, you want to start with maybe 10, but you want to make sure that those 10 are the questions with the highest volume, to create the biggest impact for your support. With clustering, you can do that instantly.

How clustering works in 3 easy steps

How does clustering work at ultimate.ai? There are literally just three steps:

I. Install and start clustering

    Step one is to install the application from Zendesk marketplace from Salesforce, Freshworks, or whatever CRM system you use. Then we automatically pull the historical conversational data and start the clustering, where the machine learning algorithms try to compare if this question or message is similar to another one. If yes, we add that into the cluster, if no, we create another cluster.

    II. Get the list of clusters

      Step two is where you already see the cluster, you get a great overview of what are the most common questions being asked, what is the distribution, so how big is the number one question. Maybe it's 10%, 15% of the total volume, which means that it's very important.

      III. Prioritize and start the automation journey

        Step three is where you prioritize. Now you have the data and the understanding to take your customer service to the next level.


        Picture that neatly organized boxes of legos, and imagine your customer service messages being just as orderly. It’s possible with clustering. Now that you understand the concepts behind clustering, and the integral role it plays in automation, you’re ready for the next step, which is to put your learnings into action. Remember: your most common customer questions probably aren’t what you think they are. Investigate, or hire a specialist to take a look. Start small, with around 10 of your most common questions, and prioritize. Now, you’re all geared up and set to start your automation journey.

         

        Similar Articles

        Discover the power of ultimate.ai

        Get started

        See ultimate.ai in action

        Watch a Demo