Lost amid all of the stories of failed chatbot implementations is this important detail: Chatbots must be conversational in order to be successful.
Chatbot systems typically fail (with failure defined as consumers asking to speak to human agents instead of dealing with a chatbot) because most of them can’t interact with people in a natural way. Most chatbots are simply “FAQ bots” — all they do is provide scripted answers to a set of frequently asked questions. They lack intelligence and are incapable of understanding the meaning of customers’ questions in order to respond accordingly. They literally just look at the customer’s words and compare them to a database of written questions and answers.
How Conversational Bots Differ
A conversational bot, however, can move beyond just answering questions to focus on intent — what the consumer is trying to do.
To understand the difference between the two types of bots, consider when booking a flight. If you ask an FAQ bot if meals are served on the flight, the bot would likely respond with a message that says something along the lines of “Meals are only served on flights of four hours or more,” along with a link to a web page explaining the meal policy. If you ask how to order a meal for your flight, you might get a link to a page where you can place an order, and if you click on the link, you would no longer be in the chatbot.
In the same scenario with a conversational chatbot, however, the artificial intelligence (AI) behind the bot would know your flight and its duration, and it might respond with “Yes, your flight includes a meal. You can choose between beef, chicken or a vegetarian option.”
See the difference? Going a step further, the conversational bot could add “Would you like to order a meal now?” Then, instead of giving you a link to a web page, it might simply ask, “Which meal would you prefer?” You would type “vegetarian,” and it would confirm that your order has been placed.
|FAQ Chatbot||Conversational Chatbot|
Consumer: “Are meals served on my flight?”
Chatbot: Link to a web page.
Consumer: <clicks on link>
Chatbot: <closes chat window>
Consumer: <reads directions, navigates to reservation page, finds reservation and goes through meal selection process>
Consumer: “Are meals served on my flight?”
Chatbot: “Yes, your flight includes a meal. You can choose between beef, chicken or a vegetarian option. Would you like to order a meal?”
Consumer: “Why yes. I’ll have the vegetarian option.”
Chatbot: “I have added the vegetarian meal to your flight.”
Related Article: AI Projects Don't Have to Break the Bank: Consider the Chatbot
The Importance of Understanding Intent
To automate conversations, a bot first needs to understand customer intent. When you call your credit card company and say, “I had a charge that didn’t go through,” a human would understand what that means. An FAQ chatbot would not. A conversational chatbot would understand the statement and the consumer’s intent behind it and could respond with “The charge didn’t go through because you’re late on your payment.”
This is not to say that FAQ bots are bad. One approach companies can take is to use an FAQ bot for rudimentary tasks and escalate to a conversational bot for complex tasks. Today when an FAQ chatbot gets stuck by a confusing (or new) input from a consumer and doesn’t know what to do next, it goes off the rails. This usually happens in response to the consumer’s most recent phrase alone, because the bot is incapable of taking the context of the entire conversation in account.
To get the interaction back on the rails, an FAQ chatbot could hand the interaction off to a conversational bot. Conversational chatbots are designed to analyze the full transcripts of interactions and understand the consumer’s intent based on the full conversation, as opposed to just trying to parse the one phrase that derailed the FAQ bot.
However, while it might be tempting to think about starting with an FAQ bot and upgrading to a conversational bot later, keep in mind that frequently asked questions represent only a small part of a typical conversation. Actual conversations are much more complex. Moreover, there are chatbots that include both FAQ and conversational capabilities within a single platform, so deploying two separate chatbots might be unnecessary.
Related Article: Want to Know the Future of Customer Experience? Ask a 2-Year-Old
6 Core Principles of Conversational Bots
Gartner predicts that by 2020, 30 percent of our interactions with technology will be through "conversations" with smart machines. A bot therefore needs to possess cognitive capabilities to understand human dialogue and navigate business systems to resolve customer issues. A number of guiding principles are important:
- Available Everywhere: An enterprise should offer assistance on all the major touch points where consumers want to interact, both modern (digital) and traditional (voice). Based on the customer’s devices and preferences, a bot should determine the best channel for the current interaction. The bot should detect presence across devices and give customers the option to switch to a more effective channel. For example, the bot can sense that a caller is also on the website, and offer to present complex information using visuals over the web rather than read it out over the phone.
- Intent Driven: Since humans may be imprecise in their communications, the ability to understand consumer intent is critical. Intent prediction goes beyond natural language processing. The platform needs to combine behavioral, transactional and external signals (such as time, weather, product availability, flight schedules) to anticipate intent or disambiguate a vague request.
- Response Personalization: Conversations are more effective when the message is meaningful and relevant, based on the individual’s demographics, preferences and interests. If there is a request for product information, the bot should personalize the response by highlighting the features and capabilities that are likely to be most useful. Furthermore, the form of a response — such as its wording, phrasing and tone — should adapt to the customer’s interaction style.
- Agent Blending: Bots need to know what they don’t know: by design, certain intents may require a human agent. Bots should also know when they are failing. If frustration is detected, the bot should escalate to a human agent before the customer abandons the chat session. In these situations, a human agent should continue the journey where the bot left off, without requiring the customer to start over. The bot can remain engaged: it can suggest responses or lookup answers to assist the agent in real time. When the conversation is back on track, the agent may re-invoke the bot to handle routine tasks, such as payment collection, account updates, or terms and conditions.
- AI Automation: Gartner states that customer self-service will increase from 50 percent of customer interactions in 2018 to 64 percent in 2022 due to advancement of AI capabilities. By learning from the end-to-end outcome of customer interactions, including actions performed by human agents, AI models can continuously improve to better predict intent and optimize conversations.
- Enterprise Ready: Bots need to integrate to CRM and other enterprise systems to apply business rules and perform transactions. As bots handle sensitive consumer data, security is paramount. These solutions will have to achieve enterprise levels of security, reliability, scalability, manageability and connectivity.
Incorporating bots into your customer experience can make a huge difference but the key is to ensure core abilities are in place that can deliver a successful and positive experience for customers.