Are Chatbots ready for mainstream adoption?

5 ways chatbots still need to evolve

It’s hard to swing a $400 juicer in Silicon Valley these days without hitting a chatbot. At the last F8, Facebook improved its chatbot game in Messenger, and everyone from Mastercard to Maroon 5 is climbing on board. Chatbots of a sort drive personal assistants like Siri on our phones and Amazon Echo in our living rooms. It’s enough to make you believe the bots are taking over.

Except they’re not, at least not yet. The technology will need to advance before “conversation” becomes a standard interface. Big brands are testing the water to gauge consumer reaction and ensure they don’t get left behind, but chatbots need to evolve in several important ways before they have a chance of being widely used. Some of the needs are obvious, like improved speech recognition, while others are more subtle, like the ability for chatbots to signal what services they have to offer.

Here are five areas where these talkative bits of AI need to improve before they really take off.

Advances in AI and Natural Language Processing

Remember the early days of the web, when pages were a sea of flashing neon and blue links? That’s where chatbots are today. If bots are to reach ubiquity, people need to be able to ask questions and place orders using natural language. Whether that’s through voice or text, users can’t be expected to master a special vocabulary. If you ask Alexa to play a song and she doesn’t understand the first time, no big deal. But if a customer can’t order a movie ticket on the first try, they’ll go elsewhere.

NLP does a reasonably good job today, but it struggles with local dialects, slang and idioms. Speech recognition programs can learn speech patterns over time -- but not if you only call a business once a year. We’re still at the early stages of human-machine interaction. Companies like Solvvy are making great strides on the Customer Experience front with automated question-answering. Its NLP + machine learning engine provides immediate self-service resolutions to complex end-user questions, by learning from prior successful agent resolutions as well as a company's knowledge base and FAQ's.

It's important to consider where to leverage the technology as it all reflects on brand image. Chatbots have to do better than simply replicating the atrocious experience of today’s automated call menus. With social media’s ability to massively amplify a bad customer interaction, businesses will want to get it right. Everything people can do today through the web and mobile apps should also be available through natural language, and we’re not there yet.

Know your customer

A huge part of any AI implementation is understanding context. Much as marketing and sales are searching for that mythical 360-degree view of the customer, chatbots need to know more about the individuals they interact with — who they are, how they got here, what they’re looking for and what they did in the past. How does that information get collected and shared among chatbots? Only then can bots reliably and consistently respond to people’s needs.

For example, Admithub began working with Georgia State University last year to build a chatbot to handle its college admissions and financial aid workflow. In the early stages, the bot helped the University process questions directed at admissions, financial aid, and student activities offices -- and it wound up increasing enrollment yield by a significant margin. Over time, the University expects that the bot will better understand the academic and financial profile of each student as they progress through the University. And by the time those students return as alumnae, the bot will know everything about them.

Machines chatting with machines

The Web is an amazingly interconnected place. Type any product into Google and you’re instantly connected to merchants that have the exact product you’re looking for in stock. Chatbots need to evolve in a similar way, so they can intelligently hand users off to each other and seamlessly take over a communication.

If I type “I want a burger” into Facebook Messenger, it should be able to broadcast that to other chatbots in a manner they understand, so another service can fill my order. On the web, this is handled through well defined REST APIs. The chatbot space has a multitude of APIs competing for attention. It needs a mature conversational API that the industry can get behind so that chatbots can interoperate.  

Illuminating what’s on offer

If I interact with an app or a web page, I can instantly see which services are available through links and other elements on the screen. Chatbots don’t have this visual language. When you talk to a chatbot, you’re going in blind. What can I ask it? What does it do? Microsoft and Amazon have worked hard to educate consumers about the capabilities of products like Cortana and Echo -- and whole articles have been written on the topic. Interacting with a chatbot for the first time, people need to know -- will this bot let me choose seats or only buy a ticket? Can I change an appointment or only make one? Am I allowed to customize this restaurant order? With no visual cues, new expectations need to be established, or a way to signal what’s on offer.

Reading emotion

Chatbots will provide infinitely better service when they can read facial features and inflections in tone to understand the emotion of the person they’re communicating with. This is partly about simple customer service -- if the user is becoming frustrated or angry, it may be time to hand the conversation off to a human. But there can also be an entire class of services, in areas like counselling or therapy, that operate based on the reactions of the user.

Chatbots have a promising future, both at work and in our personal lives, but we need to address these challenges before they can enter the mainstream. When they do, we can expect new conveniences and new experiences -- and new ways of engaging with customers. But moving too early risks alienating people before they have a chance to see the benefits.