Our AI team builds ticket categorisation algorithms for 100s of top brands. It’s our speciality.
Over the years working closely with customer service teams, we’ve seen mistakes and mishaps as well as huge triumphs.
The secret? There’s a few.
For this article, we’re going to share the good and the bad of our learnings over the years to bring you our experience-led help desk ticket categorisation best practices.
In our recent article, the importance of customer service analytics, we discussed a number of incredibly impactful reasons to understand and use support conversations in your business.
We went as far as to call them your “most valuable customer experience assets”.
This entire benefit is predicated on you having a high quality and consistent categorisation process in place.
Without a tagging taxonomy in place that:
None of it works. No CX insights. No service improvements. No reduction of costs. No revenue boost.
Building a best practice help desk ticket categorization process is the lynch pin of your customer service analytics success. It’s your foundation.
This is a quick but relevant interlude. I asked the above question to SentiSum’s Chief of Product, Kirsty Pinner, recently and her answers were directly relevant to this article.
She said, most companies:
“Get the how wrong. 1. They report on ‘reasons for contact’ through manual categorization which almost always brings inaccurate or subjective results. 2. They let customers categorize support tickets with self-reported ‘reasons for contact’. This causes an inherent distrust in the analysis, we see many support managers hesitate to share the results with other teams because they themselves don’t trust this data to back up business decisions.”
Kirsty’s point is important to note up front. However you decide to build a categorisation taxonomy, if the person or tool applying the tags gets it wrong or isn’t detailed enough, then your entire end analyses are nullified.
If you’re going to invest in support conversations as a source of company improvement—it’s worth getting right. (AI categorisation is your best bet here)
There’s lots to get through here before you start categorizing your help desk conversations. Let’s start at the beginning—your goals.
Every well done project starts with a ‘why’. The outcome you want is everything to knowing what to build.
Start by bringing other teams together to figure out the ideal end result. Jenny Dempsey, CX Manager at Apeel, said it well when I talked to her recently,
“I typically start by talking to other teams to understand what needs to be measured outside of what I/ my team want to measure. I learned early on that if I just measure what I want, I don't have access to data that other teams need.”
Most often, especially if you want to use support insights to help product, operations or CX improvements, you must rope in those teams. For example, you might want to align some of your categorizations with their most important ongoing projects—feeding them insight that makes their lives easier.
One example of this comes from an interview with Aistė Sobutienė, customer support director at Vinted. In the interview, we learned how Vinted had built customer service listening into the product development lifecycle. When a new product was launched, initial feedback was gathered from conversations so the product could be quickly iterated—a perfect use case for temporary categories to track feature feedback.
Without being on the same page as her product team in the planning stages, Aiste would likely have collected the wrong insights and they probably wouldn’t have listened.
The UK Government recently ran a ticket categorization project, they kicked off with an internal exercise with their product teams:
“In these sessions we printed off a random sample of tickets and gave each team member 5. Everyone was asked to write on Post-Its what things they’d want to capture about the ticket, for example the type of request or the date it came in.”
That’s a great way to get started, but ultimately you need to be in control of determining which tags to use.
We explain the difference between a single layered taxonomy and a multi-layered one here.
There are pros and cons to both, but ultimately if you have the resourcing then a hierarchical taxonomy is infinitely more useful. It gets you closer to the granular insights with less further root cause analyses.
Jenny Dempsey weighed-in with her best practice here, too, agreeing with us that hierarchical categorization has its benefits:
“I get very specific and create High Level Tags + Specific Reason Tags, so each ticket will have multiple tags: for example, if there is a product issue, we'll use the "product_issue" tag + a reason tag like, "shipping_delay". This helps me see how many total of one type of issue + breaking it down into smaller chunks.”
This adds complexity to your ticket categories, which brings us to point three.
“Think about your tagging structure carefully; there’s a balance between too generic and too specific” - Kirsty Pinner, Chief Product Officer at SentiSum
When building your ticket categorization taxonomy, there’s risk in going to either extreme.
In the past, customer service leaders have come to us with 400 or 500 different categories. Over time they’ve expanded with different needs and new team members, and ultimately they’ve gotten lost in the sheer volume of them. We typically streamline their taxonomy down to 30-50 tags which cover the main problems, questions and feedback in enough detail to be useful.
Tag “bloat” is a huge problem if your agents are manually categorizing conversations. Your agents probably categorize in about three seconds, if you have 500 tags they’ll likely choose the first relevant one they see and move on. Even if it’s not the best or most accurate choice.
Here’s an example of the UK Government’s (mentioned above) categories. Imagine this but 500 of them?!
That being said—if you’re too broad and high-level, using only five tags like “Complaint”, “Request”, and “Question”...what use is that to anyone?
The end result still requires a deep dive, and if you’ve got 100,000 help desk interactions a week like our customer Gousto, that’s an impossible task.
What you name your tags matters. Call them what agents would expect them to be called so they’re easy to find, easy to understand and easy to apply.
If your category for “first time login issue” is called “FTLI”, then an agent is unlikely to find it quickly.
Here are four examples of great category names:
Put the most important topic at the front (e.g Size) and the detail later on (e.g. Size_small), that makes it easier to spot.
When we dig into our clients existing tagging systems, we find that overly general tags are often a key cause of inaccurate insights.
Generally applicable categories become catch-alls, for example something like “packaging” which could apply to several different issues the customer faced with their packaging. They often get used willy-nilly by agents in a rush to move onto their next ticket.
“Avoid general tags like ‘packaging’. They tend to become a catch-all for any kind of issue, so agents apply them quickly and move on.” - Kirsty Pinner, Chief Product Officer at SentiSum.
We strongly recommend have tier 1 (high level topic tags) and tier 2 (specific subtopics tags) in place to ensure agents go deeper with their tagging.
Artificial intelligence-powered tagging doesn't run into this issue because it never tires or rushes. As long the taxonomy is set up correctly, AI will apply tags consistently every time. Read more here.
So far, we’ve been talking about topic tagging—i.e. Categorizing conversations based on the reason for contact.
But, there are TONS of different ways to categorize. For example, if you categorize by sentiment (or do a topic-based sentiment analysis) you can begin to understand which topics drive the most negativity.
If you’re going down either the manual or automated tagging route, a knowledge-base is critical documentation to get in place.
Make sure yours clearly says what each category means, where it’s supposed to be applied and why it’s important. This nuance will come in handy when new agents join or existing agents forget, or if any external teams want to sift through the data themselves.
Read our guide on 'how to create a knowledge base' here.
You’ll quickly learn what’s useful and what’s not, and new needs will arise as new bugs appear or new products are released.
It’s important that you keep an element of flexibility and allow your team to suggest new tags.
We use machine learning at SentiSum to identify a new topic that hasn’t been seen before—the speed at which it works means our customers can fight flames before they become fires.
There’s no better way to learn than by watching the innovative ways other companies categorize tickets.
Here are a few great examples:
SentiSum customers use our platform to automatically surface important topics and customer sentiments from customer service conversations. Every customer gets a custom data-driven ticket categorization taxonomy that’s continuously maintained and iterated by our team according to your needs. Our AI algorithm then applies it accurately and consistently in real-time, and the insights are presented in a highly visual and easy to deep-dive dashboard.
Book a product tour with us here to see how we can help you.