“We’re interested in using AI to understand what’s happening in customer conversations at a high level,” says Chris Hausler, Senior Data Science Manager, Zendesk. “The AI understands the patterns that emerge, and does it in a way that’s a lot more scalable than what an individual human could gain by reading through all of them.”
Reviews, support ticketing, and chat are great ways of listening to your customers as a company, Hausler says. It’s how to stay on top of customer sentiment around your product. Reviews are the external-facing piece of that, where customers can share what they feel about a product with other customers who are going to potentially buy that product.
Internal and external signals
“Customer reviews are hugely influential,” he says. “In principle, it’s the non-biased signal that people are looking for when they’re going to potentially purchase a product. Understanding how other people who’ve already purchased a product feel about it, what they think was good, what they think was negative, is a really important signal for those potential customers in making a purchasing decision.”
Zendesk’s Gather product, which is a forum that allows customers of a company to discuss products, ask for help, give feedback on products in a publicly moderated sort of fashion, is a major way companies collect customer sentiment signals and determine patterns of pain, patterns of need, and collect requests.
“The forum is not as explicit a signal as you would expect from a four or five-star rating out of five,” he says “but there’s still a huge amount of information to be gleaned there that you can miss if you’re just focusing specifically on the rating-type feedback.”
But whether it’s reviews, comments in a forum, or support requests, the challenge from a technical perspective is that it’s quite hard at scale to be able to take all of that text that people are writing and understand the core themes. AI is key.
The role of AI
“We use AI to dive into those pieces of text and scale and understand the patterns that drive those comments,” says Hauser. “As a company, it’s important to be able to do that and step back from the individual feedback and understand the broad themes that people are talking about. AI is a way of being able to find those patterns and provide them to a company so they can then understand what people like and dislike, and take that into account as they plan future iterations of their product.”
Companies can also look at their support requests and use AI to find groups or topics within them to summarize their customers’ pain points and bubble them up to administrators to help refine and iterate on their knowledge base content, making sure that they have the key pieces of information available that their customers are looking for without having to have an agent involved.
Hausler has found that the feedback and the topics that have been bubbled up for companies are often surprising – they had not realized that customers were talking about some of the issues or areas that have come to light, or hadn’t known they weren’t providing sufficient information.
How it works
Under the hood, the AI is quite complex, he says. Synthesizing information is something that humans are intuitively very good at doing. He uses the example of how people easily categorize books; give anyone a group of books on different topics and ask them to organize those books into groups, and then explain what each of the groups are about. Almost anyone could complete that task – but it’s actually quite challenging for a machine. There are quite a number of steps required to make that happen and make it happen at scale.
At a really high level, the first thing is to be able to work out a way of understanding how similar two pieces of text are, which requires deep learning to map each piece of text, effectively, as a set of numbers that can be compared. Working out how the text forms groups requires clustering, or determining which pieces of text are similar to each other and dissimilar to other pieces of text. In the book analogy, it would be grouping science fiction over here, and the books on 19th century history over in the corner. In the AI world, that’s quite math intensive.
For support tickets, they’ve refined the process of grouping tickets and then summarizing them in a way that’s accessible by the administrator — a list of topics, the volume of tickets in each group, and so on, so that they can make sense of where the issues are emerging and then act on them.
Starting with the customer
It’s all about being obsessed with the customer, or being customer-centric, Hausler says. That has to be the starting point for any of this.
“Being able to glean all of this information is only important if you’re coming at it with the focus of wanting to take that information and do something about it,” he explains. “Being able to take a step back from the individual reviews or support requests you’re getting and understand, as a whole, what are the core things that people are talking about in your product? What are the core drivers that make them happy or not happy, and be able to take that to adapt to how you’re running a business and provide a better service for your customers.”
To learn more about how staying on top of customer communication can drive company success, how to tap into those dialogues, identify patterns of concern or areas of improvement, and leverage AI to make it happen at scale, don’t miss this VB Live event.
You’ll learn:
- How to find themes in your customer reviews so that you can fix the real pain points instead of putting “band-aids” on each unhappy customer
- Steps to guide your business decisions around what your customers say they want, not what you think they want
- The importance of addressing your most common negative issues first to see an immediate change in your customer satisfaction level
- The concept of always improving your customer experience – searching for trends in your good reviews to turn them into great reviews
Speakers:
- Chris Hausler, Senior Data Science Manager, Zendesk
- Ramin Vatanparast, Chief Product Officer, TrustPilot
Author: VB Staff
Source: Venturebeat
