The Secret Power of Data Science in Customer Support

content online focuses on how it can be applied in Product or Marketing — the two most common fields where data scientists create great value. However, working at a startup, I’ve had to work with many more functions outside these two. Data exists across the company, and the reality is that every department can benefit from data science and Analytics to improve efficiency and drive business value. In this article, I am going to discuss one of those less-covered topics — data science for the Customer Support (CX) team.

I remember the first time I was pulled into a meeting with the CX team, I was completely clueless. I didn’t know what to expect or how data could actually help them. But now I have worked with the team for over three years as their Data Science partner, from the early days when we barely had any data reporting to now, when we are deeply embedded in the function and support data-driven decisions. In the sections below, let me go through the common data science use cases in CX.

---

1. Metrics Tracking

Before you can improve anything, you have to measure it — and CX is no exception. Building metrics is also a good way to establish trust with your stakeholders. 

For CX specifically, some common metrics include:

• SLA (Service Level Agreement): This is the commitment or target for how quickly the customer support team responds to customer contacts. For example, “respond to all chats within 3 minutes.” It is critical to monitor whether the team always complies with the SLA. It is typically measured as the percentage of support interactions that meet this goal. 
• TTR (Time to Resolution): SLA cares about whether each interaction was done in a timely manner, while TTR measures the total time it takes to resolve a support ticket — including all the back and forth. Imagine you, as a user, reached out to customer support via email for a product question. They responded quickly every time you messaged them, but none of the replies actually solved the question. In this case, SLA would look good, but TTR would be long. That’s why we need both to complete the story.
• FCR (First Contact Resolution): Ideally, the customer will be provided with what exactly they are looking for in the very first conversation. Therefore, FCR is designed to measure the percentage of support tickets that are resolved without needing follow-ups. Naturally, a low FCR is correlated with a high TTR.
• CSAT (Customer Satisfaction Score): The above metrics are all internal measures of how quickly we get back to our customers and solve the issues, while CSAT is a direct external measure of how satisfied customers are with the support they received. It is often captured via a survey after a support ticket is resolved, with a question like “How satisfied were you with the support you received?” (score 1 to 5). 
• Contact Rate: We care about the quality of the service, but it is equally important to understand how many support cases are generated. A great way to normalize the case volume is to calculate the Contact Rate as the number of cases / number of active customers. This tells us how often customers encounter issues and need help, so it is also a measure of product friction. 

Of course, there are many more metrics we have built for the CX team, but the above metrics should give you a good first glimpse into what data matters to the CX team. They, of course, are organized and presented in dashboards so the team can monitor the performance and dive into certain case types, teams, or customer segments. At my company, the data team also co-hosts a weekly metrics review meeting to spot trends, surface insights, and drive discussions. 

Now that we have all these metrics, how shall we utilize them to drive changes? That’s where the real power of data science comes in. See the following use cases. 

2. Workforce Management

Each customer support interaction results in labor costs as well as technology costs, overhead costs, and other operational costs that come with it. Therefore, it is critical to accurately monitor capacity and forecast future support demand for staffing and planning.

The data team can provide lots of value here:

• Forecasting contact volume: This is a complex but high-impact task. It first requires cross-functional collaboration to get the right assumption of customer growth projections and adjust the contact rate expectation given product launches and improvements. Then, data scientists can utilize data toolkits like time series models to bake in all the assumptions and predict the support case volume. 
• Capacity planning: Once we get a good prediction of contact volume, the next question is how many support agents we will need to maintain a good level of service. This requires scenario simulation of agent performance and availability, and optimization of the agent shift schedules to ensure we meet SLAs without overstaffing.   

3. Process Improvements

Data is not only helpful to track the team performance, but it can also drive real process improvements. Just to give you a few examples that I have seen:

• TTR analysis: TTR is just a random large number without making sense of it. The data team can analyze TTR to identify drivers of long resolution time and use that to inform process improvements. For example, if the onboarding-related cases often take a longer time with many back-and-forths, this could imply that the CX team needs more training regarding the current onboarding process, or the onboarding flow is over-complicated, so customers constantly find it confusing. If the cases coming from email usually have a long time to resolution with a low CSAT, maybe we should allocate more resources to answer the email queue to speed up the responses, or provide better tooling support to help agents draft their emails. 
• Support tiering strategy: Not all customers are of equal value to a business. Therefore, a common practice is to create support tiers among customers and prioritize the contacts from top-tier customers. The data team can help come up with the tiering system based on customer value and monitor the effectiveness over time.
• A/B testing of support flow: Where should we put the live chat button? How to make the support center more discoverable for customers? Is a certain auto-reply email format better than another? A/B testing method helps us answer these support flow design questions. 
• Self-service enhancements: The ideal world of customer support is no human support needed 🙂 Though this is nearly impossible to reach, the data team can help to get closer. For example, we looked at what kind of questions users failed to resolve via the help center. This informs what new topics should be added to the help articles and how the help center search function should be improved. 
• Chatbot improvements: Chatbot is a common tool to answer customers’ questions without routing to real agents. Especially in this AI era, we have seen significant improvements in chatbot quality and availability. Our data team has played a critical role in two rounds of chatbot vendor evaluation with the CX team — setting up the data pipeline, A/B testing of different chatbot options, evaluating chatbot performance, identifying the low-performing contact categories, and helping fine-tune the bots to achieve a better chatbot containment rate. 

4. Customer Feedback Analysis

Last but not least, support contacts generate a great amount of text data — they come directly from the customers and can be used to understand customer pain points and product gaps. 

• Case categorization: Support cases can be categorized manually by the CX team or with a rule-based framework, but the data team can help to automate this step, especially with AI’s power today. With simple prompt engineering, most LLMs today can categorize each case based on your product context with decent accuracy. 
• Text analysis: Except from categorization, AI can take the whole case transcripts to summarize and identify the customer pain points. My team collaborated with the engineers to build an internal AI product called “Voice of the Customers” that processes all case details through LLM and surfaces the most common customer complaints in each product area. This is a perfect opportunity to bring CX insights to the whole company and close the feedback loop with product and marketing. We have seen it being used actively in product roadmapping. 

---

Working with the CX team has been an unexpected but rewarding part of my data science journey. From tracking team performance, supporting capacity planning, to optimizing internal processes, and improving customer experiences, data science can really transform how the customer support team operates.