What could you accomplish if you could improve the quality of your decisions?
You could avoid getting this haircut.
You could avoid this shortcut.
And you could improve your customer support.
“Studies strongly indicate marked performance differences in those organizations that embrace the opportunities around the broad space of analytics,” wrote Bob Picciano a Software Sales GM at IBM. “Our own study, conducted jointly by IBM and MIT, showed that organizations that focus on analytics significantly outperform their segment peers on the key business metrics of growth, earnings, and performance.”
So what is data science? This short guide will introduce you to the goals of data science, as well as the four steps to using it to improve your customer support.
The Goal of Data Science
“Data Science is the art of turning data into actions,” according to the Booz Allen 2015 Field Guide to Data Science. The goal: better decision-making.
To get there, you need what Booz Allen calls a “data product.” This is the recommended action. It’s based on data, but you don’t need to see the data to understand and implement the recommendation.
The magic happens when you turn data into recommendations and recommendations into actions.
Examples of Data Science
One of the simplest examples of data science is radio frequency identification (RFID) tagging. Today, companies can easily and precisely locate every item they’re responsible for, from products to livestock to conference attendees to luggage. And they can track these items along their paths. RFID tagging makes it easy to make sure food doesn’t get too warm in transit, make sure a bridge is holding strong, and make sure railroad tracks haven’t expanded beyond what’s safe in the heat.
Other examples, from Booz Allen:
- Movie Recommendations
- Weather Forecasts
- Stock Market Predictions
- Production Process Improvements
- Health Diagnosis
- Flu Trend Predictions
- Targeted Advertising
Steps to Data Science
Step one: Decide what your most pressing questions are
What are the questions that, if answered, would provide the most value to your organization?
One reason you want to identify your questions before you get into your data is that big data by itself is overwhelming. It’s like going into a new Super Walmart with no idea what you need to buy. You’re going to wander around aimlessly, get lost, and feel overwhelmed. And you may or may not walk out with anything valuable.
For example, ridesharing company Lyft figured out that it could be very profitable for them to know which customers they were most in danger of losing before they lost them.
Step two: Identify the pressing questions data can answer
Not all these pressing questions can be answered with data. So you need to isolate the ones that can.
Why customers leave Lyft can’t be answered with data as easily as which customers are about to leave, because “why” in this instance is a qualitative question, but “which” is a quantitative question.
As WordPress.com Team Lead Simon Ouderkirk put it for Help Scout, “Leveraging your existing customer support data to unlock the value present in your support unit begins with asking the right questions.”
Examples of good questions for data science from Booz Allen:
- Which of my products should I advertise more heavily to increase profit?
- How can I improve my compliance program, while reducing costs?
- What manufacturing process change will allow me to build a better product?
For support teams, Ouderkirk recommends starting with your assumptions about your customer base. “Identify the big, untested beliefs that power your support team.”
Here are some assumptions about what customers of Automattic (the parent company for WordPress.com) might want:
- Plugins for their sites
- Support to be in English
- All replies to come from the same person, even if it takes longer to get a response
Then you turn those assumptions into “Is it true?” questions. Examples:
- Is it true our customers want plugins for their sites?
- Is it true our customers speak English first and every other language is a distant second?
- Is it true our customers prefer replies from the same person, even if it takes longer to get them?
Step three: Extract and analyze the data
I spoke with @bht on Support Driven Chat about about Lyft’s data science win.
After asking why customer leave Lyft, the team decided to combine customer support email data with data about payments to its drivers to try to identify a pattern. It discovered that the fifth email was often the last email from a particular type of customer about a particular topic before that customer “would fall off the wagon and we’d lose them as a customer.”
Initially, the Lyft team team crunched the data manually, but then it found a way to combine raw data from desk.com about its customers with internal data on the same customers to correlate the two and verify its hypothesis.
Another example of a data-driven insight comes from @davedyson on Support Driven Chat. “A high average-subsequent-response time could be indicative of agents taking on more new tickets than they can adequately manage,” he wrote.
Your analysis should result in a finding. That finding must answer pressing business questions in such a way as decision makers can understand the answer without much background.
For Ouderkirk, the key to finding the data that can answer his questions lies in re-asking the questions in light of customer behavior. Ask yourself, “What measurable behavior would our customers engage in if this belief were true?” So if, for example, Automattic’s customers want plugins for their sites, what behavior would they be engaging in? Well, they might be searching the knowledge base for “plugins.” That’s easy enough to answer with data.
Step four: Make a recommendation
Ask yourself, “What is most important to this decision maker today?”
Then, use data to show how the issue you’re championing can have a direct impact on what matters to them. Are you in a high-growth startup where “Monthly Active Users” is your most important metric? Use data to show how support can help move that needle once they have the resources they need. Are you in a mature company, struggling with turbulent retention rates? Use data to show support’s impact on customer retention.
Learning that the fifth email was the drop-off point for Lyft customers “helped us prove to Project Managers and Engineering that it needed to be fixed,” @bht told me. Now the team can point to data showing a particular issue was causing the company to lose customers even before the fifth email, “but it was an almost guarantee at the fifth.”
Another example of a recommendation you could make based on data science comes from MarketingProfs. According to a recent Bizrate survey of 100,000 shoppers, their biggest complaint was having to pinch and zoom their screen to click a button when shopping on their phones. Shoppers often still ended up clicking the wrong link after all that — obviously irritating. The fix is easy, just separate the links more and pad the buttons with more pixels to make them easier to hit.
Bizrate got this information from a survey, but they also could have gleaned it from their analytics. First, you need to track which actions people commonly take right before leaving your site. Examine any actions other than a purchase or other conversion to find friction. Then test ways to smooth it out.
MarketingProfs: “Abandoned shopping carts, bad reviews, unused features, high return rates, and a bad reputation in social media are all examples of consumers voicing their opinions.” Data science helps you listen to the customer’s voice.
There might be only four steps, but data science isn’t easy by any stretch of the imagination. Don’t be discouraged. Booz Allen Hamilton: “We have observed very few organizations actually operating the highest levels of maturity, the Predict and Advise stages.”
Are you using data science in your customer support department? Why or why not? Let me know in the comments!
Looking for Customer Service software? Check out Capterra's list of the best Customer Service software solutions.