Zendesk Doubles Down on Machine Learning With Guide

By Rob Marvin Features PCmag

Zendesk keeps getting smarter. The helpdesk provider today announced Zendesk Guide, a knowledge base product imbued with deep learning algorithms to provide more intelligent, contextual customer service responses.

Continue Reading Below

Zendesk is in the midst of a multifaceted re-brand, building around its flagship Zendesk Support platform. The company now offers Zendesk Chat, Talk, and Message, which are standalone products for live chat, call center software, and in-app customer support messaging (within Facebook Messenger and Twitter), respectively. The newly announced Zendesk Guide is a revamp of the company's knowledge base (previously called Help Center); it boasts new features and functionality centered around machine learning (ML).

The growing emphasis on artificial intelligence (AI) is a more subtle dimension to Zendesk's transition beneath the growing product family. Last year, the company added ML and predictive analytics capabilities to the platform with its Satisfaction Prediction feature. Zendesk Guide builds on that strategy with a new "Knowledge Capture" app for agents as well as "Answer Bot," an AI bot that uses ML algorithms to provide more relevant responses, learning over time within the self-service, automated experience of Zendesk Guide.

"It's been a maturing progression, moving from satisfaction prediction—which was really grounded in traditional techniques like logistical regression—to the more cutting-edge deep learning and natural language processing [NLP] we're using in Guide," said Jason Maynard, General Manager of Zendesk Guide.

"This model is interesting because it's trained on a global level across every customer's data and solves two problems: cold start, where a customer who just got onto Zendesk can now start getting answers right away, and it also means the model itself will tune itself to be more effective. Because we're not limited to training [the AI] customer by customer, it allows a lot of SMBs with smaller customer bases to still take advantage of machine learning."

Inside Zendesk Guide

Continue Reading Below

Zendesk Guide is a new product that comes in two plans: the free "Guide Lite" plan and the "Guide Professional" plan, which costs $15 per agent per month. Existing Zendesk Support customers will be grandfathered into one of the plans. Zendesk Essentials customers will have access to Guide Lite, while current Zendesk Professional, Enterprise, and Elite subscribers will be grandfathered into the premium Guide functionality. Both plans include Guide knowledge base access (the re-branded Help Center). However, only Guide Professional users can access the two primary features of the new offering: Answer Bot and the Knowledge Capture app.

The Knowledge Capture app is a simple, crowdsourced way for agents to naturally expand a company's knowledge base. Through a couple of simple steps in the app, an agent in Zendesk Support can publish a new knowledge base article directly to the Guide's Help center, web widget, or mobile app. It's a simple feature but one Maynard said incorporates the principles of IT Service Management (ITSM) in a more straightforward and intuitive way, giving agents a proactive workflow to fill in larger knowledge gaps.

"Looking at the principles of knowledge-centered support, a well-established ITSM principle is how to get agents involved in knowledge creation," said Maynard. "ITSM works great for customers who want that whole process but it can be a bit heavy for some. Launching the Knowledge Capture app—which is available within Guide to agents within the ticket interface—takes some of the principles of ITSM and lets agents apply that within a ticket. An agent resolves a ticket, gets feedback, puts that into a template, and submits it to the content team as a baseline to build an article."

Answer Bot is similarly straightforward from a user perspective. The bot responds to customers with recommended articles and gives users the option to solve tickets before they even get to an agent. At the algorithm level, the AI is learning as it goes. As customers write queries, Answer Bot's deep learning and NLP model analyzes the semantics of the question to find relevant knowledge base articles and automatically surfaces the answers in a chat-based interface.

The self-service experience lets customers close their request if the answer was helpful or provide further feedback to teach Answer Bot or to refine the query. There are also some additional feature flourishes, such as the ability to converse with Answer Bot by using emojis.

Maynard said Zendesk first started experimenting with ML in 2013. Now that Zendesk has a couple of years' worth of deep learning experience and training ML bots under its belt, Answer Bot has evolved from both a data analysis and user interaction perspective.

"Answer Bot went through a lot of iterations. We found that tweaks in the user experience were driving up drop-offs and deflection rates," said Maynard. "We realized the link we were sending put users behind an authentication wall. Now when a customer emails in, we confirm their identity, and then send back an authenticated token with the link to interact directly with Answer Bot. That little tweak significantly reduced the friction between customers and the information they need."

"So it's a two-pronged approach," added Maynard. "The machine learning model and data science has to be right, but those improvements to the user experience end up having just as big an impact in driving more value."

This article originally appeared on PCMag.com.