To Automate or Not to Automate: The Benefits and Limitations of AI Chatbots
Brands love chatbots. According to Juniper Research, chatbot technology will save businesses $8B annually by 2022. And consumers want to love them, too. But customers don’t always have such warm fuzzy feelings for chatbots, particularly those powered by artificial intelligence (AI chatbots).
For one thing, a lot of AI chatbots are only as good as their initial programming, mostly limited to processing key phrases through natural language processing (NLP) and returning obvious responses:
Customer: I have a question about my shipment.
Chatbot: How can I help you with your shipment?
Customer: When will it arrive?
Chatbot: Do you want to know when your shipment will arrive?
And so on in a maddening loop that sometimes, but not always, goes somewhere.
For AI chatbots, NLP technology is not yet in a place where it can accurately simulate human interaction. No chatbot has come anywhere near passing the Turing test, and many of today’s AI chatbots require an unrealistic amount of training data to get to a truly usable place in a practical setting — which is why AI chatbots are only effective in limited situations.
This doesn’t mean brands should ignore chatbots, though. It simply means brands should be purposeful in how they use AI and chatbots together. Here are some best practice recommendations for using AI and non-AI chatbots within a customer service model.
AI for answerbot and classification
Using AI for general-purpose problem-solving is not practical. But AI within chat excels when used for a very specific purpose. The current most effective and practical uses of AI chatbots are for knowledge article suggestions and self-service.
Answer bots with AI chatbots
Answer bots are AI chatbots that can be used within a web or in-app messaging format to automatically respond to users’ customer service queries by directing users to knowledge base articles. For instance, an answer bot might handle a question like this:
The answer bot takes keywords from the customer’s query and matches them up with tagged FAQs. The better your FAQs are tagged, the more accurate the answer bot’s responses become.
AI-powered issue classification is the mechanism by which chatbots get assigned issues. Manual categorization is resource-intensive for many organizations, and users are not always great at knowing which department they need to speak to in the first place if their issue falls into a gray area. Automation via AI classification provides a smarter alternative.
To train an AI classification engine, the engine is fed a sampling of issues across various categories. Once the classification engine has reached a reasonable level of accuracy after learning from the existing categorization, it can then start categorizing issues so they are directed to the most appropriate bot or agent .
As an alternative to AI chatbots, decision-tree bots prove to be more functional. This type of bot is designed to lead customers down a predefined workflow based towards their resolution. This non-AI chatbot typically presents the customer with multiple-choice questions such as:
What can I help you with today?
The choice the user makes determines the next information or options the user sees. It’s a very controlled, logical flow.
Decision-tree bots can be used for information collection, full issue automation, CSAT collection, and more. Decision-tree bots can hand off to an agent when necessary, so the customer’s experience is fluid and seamless within an intuitive messaging interface. The customer does not have to shift gears or channels to be “transferred” to an agent.
Decision Tree and AI Chatbot Best Practices
Regardless of the type of chatbot you’re engaging, there are some common best practices to consider.
Penn State research reveals that people get easily annoyed by chatbots with humanlike features that aren’t effectively interactive. Customers don’t want even their AI chatbots to seem human. “When there are high anthropomorphic visual cues, it may set up your expectations for high interactivity — and when the chatbot doesn’t deliver that — it may leave you disappointed,” says S. Shyam Sundar, James P. Jimirro Professor of Media Effects, co-director of the Media Effects Research Laboratory and affiliate of Penn State’s Institute for CyberScience (ICS).
The idea of chatbots isn’t to trick users into thinking they’re talking to a real agent when they’re actually interacting with a machine. Instead, the best use of chatbots is to solve the customer’s problem, quickly, and pass that customer along to an agent when it can’t. Transparency is more obvious in a chatbot driven by a decision tree.
Make the Transition to a Human Agent Fluid
Neither decision-tree nor AI chatbots can resolve every customer query. Ad hoc and unusual requests still may require human assistance.
When it becomes necessary or useful to connect the user with a human agent, that transition needs to be both obvious and smooth. The user should know that they’re now talking to a person, but they shouldn’t have to repeat their details to that person. For this reason, it’s critical that your use of chatbots be integrated well within your customer service platform, and that agents have access to the customer’s entire conversation history.
Identify Clear Use Cases
The key to effective chatbot usage is to identify the use cases that automation can handle and those that will benefit from the human touch. Issues that tend to repeat themselves among customers are ripe for automation. For instance, users often contact customer support to update their accounts, change or retrieve their password, or initiate a return.
The use cases that are ideally solved by chatbots are the ones that are common and require a specific, routine set of steps to remedy. Chatbots can easily follow the same protocol human agents can when it comes to unlocking accounts, for instance.
Make the Best Use of All Your Chatbots
The key to effective use of AI and non-AI chatbots is to assist, not frustrate, your customers. This is best accomplished with a smart mix of automation, AI, and human customer service. Helpshift enables smart use of chatbots with guided workflows that put decision tree bots in place and guide the customer through a logical conversation flow. Helpshift’s chatbots integrate with any API-enabled system, so logic can be built in for information validation to fully automate many customer issues.
And Helpshift makes it seamless to begin customer service conversations with a chatbot, then transition in a friction-free way to a human agent or another type of bot, within the same interface, or even across platforms.