4 Tips for Building Tagging Systems

Introduction

Most customer service departments are born from a rushed necessity and often with no planning. It is easy to overlook the value of building data structures and tagging systems because the first order of business is to provide basic service to customers. There is also, most likely, a backlog that is growing out of control. In our world of big data you will inevitably be asked about the numbers. If you haven’t tended to these issues, you will not have good answers. Here are four things to keep in mind as you try to solve the problem of reliable contact issue data for customer understanding and problem solving.

Start Basic

Starting with the core fundamentals creates a simple, common base of understanding. Remember that all of your agents (and/or AI) will have to correctly apply these as well, so simplicity will help with adoption and accuracy. Give a lot of thought to the basics because they will become the core of your data and how knowledge is categorized by people in your organization. Any mistakes here will cause significant confusion at scale. I don’t recommend going to any deeper granularity beyond three layers because neither AI nor humans are great at it. Here’s a image to show what I mean:

Generic Tagging System for One Game

Avoid the Temptation to Over-engineer Your Tagging System

People have a tendency to overengineer systems once they’re in place. Avoid this as much as you can manage because it will add unnecessary cognitive load to your agent workflow. Bloated tagging systems ensure that many tags will remain unused. It also tends to muddy your data due to increased inaccuracy.

Maintain Your Tagging System and Data

In the same way that development teams have to maintain code, customer service teams should maintain their tagging structures. They are directly linked to your data and if not built and maintained properly will cause problems in your ability to accurately identify and solve problems. It will probably prove very difficult to prevent tag bloat, especially the longer you operate, so dedicating regular times for cleanup and maintenance can help your efforts to keep things simple and useful.

For those of you running operations, remember - 95% compliance, 80% accuracy for tagging.

Beware of the Grand Promises of AI for Auto-tagging

Most software being sold for auto-tagging and categorization are really good in pockets of data, but terrible for others. Sales will always emphasize the positive numbers around their best-case scenarios, so be aware. You should always check to see how their technology works with your actual data before purchasing.

A bigger problem, however, is the type of AI models used, most of which require heavy human training. If you do not have the appropriate staff to operate, train, and maintain some types of software, you’ll be burning money. I have found that these can be helpful for some high level investigation, as part of few and very well understood automated processes, possible liveops identification of new trends, but aren’t very useful for much else. Future AI developments seem much more promising than most of what’s being sold today.

Conclusion

1. Keep your structure simple, especially at the top.

2. Avoid the logic, “if I have a tag, I will get the data I want”, which should help prevent bloat.

3. Create processes that force your team to review tag performance and usage regularly. Optimize for utility and accuracy.

4. Understand what kind of AI you’re using/purchasing and what that means for long-term operation and improvement. Make wise investments.

The best solutions combine AI/ML technology with human understanding in processes that improve accuracy of data over time, and maintain it at high levels.