How bots bring home the bacon for customer support
Anyone who has seen the futuristic, leaving-Planet-Earth flick Passengers would agree the movie offers the opportunity for an energetic debate on Artificial Intelligence—just where exactly should one draw the line?
In the movie, a robot bartender, Arthur (portrayed by Michael Sheen), gracefully fulfills his duty of fixing drinks aboard the Avalon. But he grievously falls short of the bar at a key human trait–discretion. In a conversation, he lays out the truth threadbare at a point a human bartender would have decided to say less, causing a rift between the only two humans awake on the craft. The ramifications of that rift are, nevertheless, the stuff of transcendental romance.
Arthur’s limitations hold up a mirror, too. They show where humans need replacement, where they need assistance, and where they are better left to their own devices.
For software aficionados, AI has turned into an indispensable tool in enabling humans devote their workforce toward more productive tasks. Consider for a moment the varied forms of software-defined bots helping consumers: from airlines to cinemas to food delivery services. While consumer-facing bots have been around for a while now, the widespread digital adoption over the last decade, accelerated in the post-pandemic world, makes bots the first agent of customer interaction for many brands, across many business segments.
Where there was a weary-eyed support exec, there now exists—a decision tree. And, as more consumers interact with chatbots, the more numerous its resource pool of scenarios and, thereby, the wider its branches.
From IVRs on chat to productivity tool
The query-addressing bots, which began as interactive, chat versions of the Interactive Voice Response (IVR) systems, have come a long way. The underlying technology has evolved for decades, witnessing applications in myriad fields oriented around businesses and individuals. The bots assisting companies that conduct their business predominantly over the Internet originally began as tools for automating FAQs. Instead of answering customers’ questions one by one, making them sift through a laundry list of “frequently asked questions”, companies made FAQs interactive and automated through what are called “answer bots”. The answer bots were just simple manifestations of software-controlled logic: If this is the question, then this is the answer.
The push for bots’ adoption came from the customer’s end, too. And smartphones played a big part in this. Over the last decade, a whole range of digital-first companies pushed the mobile application as their numero uno interface. Remember the big leaps by fashion e-tailers and cab-hailing apps to go app-only about six years ago? Such transitions provided the necessary impetus for bots to become must-have app features from merely being nice-to-have digital assistants.
At Freshworks, product folks say the bot journey has seen value getting unlocked at several points over the years. From being platforms for one-to-many information dissemination, chatbots became a superb tool for proactive customer service. For industries such as food delivery, chatbots have already found rock-solid use cases in process automation.
Natarajan Chandrasekaran, Senior Manager of Product Marketing at Freshworks, says food delivery apps address a large portion of queries for which the bot pretty much knows how the conversation is going to go. Take a listen.
Behind the bots’ seamless query resolution sits a technology that enables machines to understand humans, made more and more efficient through continuous feedback. We caught up with two of Freshworks’ product management folks, Vineet Gupta and Shubam Goyal, to get under the hood. The technology moved from “word tokenizers”, which are basically vectors assigned to word tokens, to the Transformer architecture for enabling bots to understand natural language. This transition has deepened the “understanding” among bots about the intent of a user.
The Tech that makes it hum
Here’s a brief snapshot of the distance Freshworks has travelled in making bots more intelligent for customers, and where the tech is headed:
- A robust intent detection platform which has a few pre-built intents like general intents (greetings, acknowledgements, frustration detection, etc.). Includes support specific intents (transferring to agents, creating tickets) available out of box and customers can add their own intents as well.
- All solution articles are leveraged to provide additional intents for bots to detect and leverage in-query resolution.
- Currently supported languages: English, German, Dutch, Portuguese, French, Spanish, and Italian.
- On the roadmap:
- NLP capabilities for other languages to be added so as to expand the set of languages supported by our bots.
- Continuously monitor and improve relevance on default and custom intents.
- Identify entities present in the end-user’s chat message to improve intent identification and simplify triggering of flows.
In the words of Vineet Gupta, the leap in intent detection and entity classification technologies could well provide the impetus for another round of innovations in the field of chatbots: “In terms of technology, what we have today is much better than what was available some two-three years ago. Earlier, it was just keyword based; later the comprehension was more semantic-based but the comprehension was still dependent on words, their adjacent words and synonyms, etc. Now, we have the technology to enable comprehension at a more holistic level.”
The next frontier: sentiment analysis
“This customer is spitting fire. Can we address his grievance asap?” That’s sage advice from a robot that can pick up the tone of voice. The next frontier to scale for chatbot builders in the consumer Internet space could well be the accurate detection of the human mood. Already, chatbots classify users on “degrees of ire” by detecting certain words, but the challenge has been to glean granular, actionable data on customer mood and satisfaction levels over a period of time. In an industry like food delivery where such insights could lead to quicker response and remedy, the opportunities appear endless. As we step further into 2021 after a roller-coaster year, AI seems poised to bring home the bacon with a giant leap in chatbot-led customer support.
Cover image and infographics: Vignesh Rajan
