4 tips to improve your chatbot user experience
Many companies employ automated chatbots as part of the customer service offering. A chatbot is great because it can be exceptionally flexible in conversational interaction. They are cheaper to “employ” than human customer service representatives and are available 24 hours a day and 7 days a week. Further, they can be deployed over several different communication channels, websites, email, SMS text messages, and social media just to name a few.
Some things to consider when you are using a chatbot are:
- “Who” do you want your customer to be chatting with?
- What transactions or interactions do you want to support?
- Who is your main audience and what is their motivation for interacting via a chatbot?
- How can the chatbot lead to other customer service options?
Each customer’s experience and expectations can be different when interacting with an AI bot. They must be accounted for as you design the user experience and craft how your AI bot will converse with your valuable customers.
Based on these considerations, here are four tips to improve the user experience of your chatbot:
Clearly identify “who” your bot is
The technology that gives your bot its voice and its vocabulary is a form of machine learning called Language Understanding and builds natural language into your AI Service. We at Neal Analytics partner with Microsoft and their corresponding Cognitive Service called the LUIS Service.
Microsoft’s LUIS Service is set up, and trained extensively so that it can appear to an unknowing customer to be a human counterpart on the other end of the chat. This says a lot about Microsoft’s LUIS Service, but from a customer experience perspective, if the customer starts to sense they are communicating with a bot, it can become frustrating and prematurely cause them to abandon the experience.
Our recommendation here is to be clear with your customer that they are communicating with an AI bot. Name the bot after your product or service like the Neal Analytics “Neal the Bot” (not a real thing) or the “Juniper Bot” or even “Robbie the Customer Service Bot.” Naming your bot in such a way lets your customer know they are talking to a computer and therefore sets their expectation early on that they may have to be more specific or even more patient than if they were interacting with a human.
De-humanizing the bot doesn’t change how you approach your LUIS model or your conversational strategy, but it does help your customer understand that they are interacting with artificial intelligence and not a person. This makes the entire transaction less frustrating than if the interaction is presented as talking to a customer service representative.
Keep the transaction simple
Depending on the type of transactions you support with your AI bot, you should anticipate what your customer will want out of their interaction. A great place to start is by analyzing the most frequent customer service calls and what they are trying to accomplish. Offering up these details or information early and unprompted should be something you consider.
For example, if your bot provides customer service for self-service account management on a stored value account, the most common piece of information your customer will want to know is their current account balance. In a recent SMS chatbot deployment Neal Analytics built, we did just that. The bot detects the inbound telephone number of the customer and checks it against the account database if that phone number returns and the phone number of record on the account, the chatbot will return a greeting something to the effect of “Hello Mr. Smith, you texted me from 555-555-5555, your account balance is $49.57. Can I help you with a transaction?”
You may think that there is the opportunity to accidentally interact with the wrong account, but the next immediate steps in the interaction are verifying that the customer is indeed Mr. Smith and authenticating him as an authorized user on the account.
However, the most important thing here is that by doing the balance inquiry as an assumption at the beginning of the interaction, we have eliminated extraneous potential steps of pre-authentication, then presenting the balance then moving on to another type of transaction. Instead, now armed with his account balance, Mr. Smith can decide if he needs to do another transaction or has a sufficient account balance to go on with his business.
Be prescriptive in your interaction
While your LUIS Model should be robust enough to decipher your customer’s intents while interpreting the words your customer uses, being prescriptive is the most direct and efficient way to understand what the customer wants and deliver on their needs. Using account authentication as an example, your dialog could be “Let’s log into your account. To do this, I will need two pieces of information from the following list: Email address, Account number, Password, and home Address” From here the customer may provide all of the information in a single message or may provide two pieces of information or may provide one piece of information without indicating what that information is.
The prescriptive way to do the same transaction is as follows: “Let’s log into your account. To do so you will need to provide us with two pieces of information from the following list: 1. Email address 2. Account number 3. Password 4. Home address. Which would you like to start with? Please type 1, 2, 3 or 4.”
Being prescriptive allows the customer to efficiently provide the proper information without introducing confusion by being too open-ended. Your customer will be thankful that they have a straightforward transaction flow and that their time is not wasted with misinterpretations of their answers.
Allow conversations to fail… gracefully
Even with the best LUIS Model and the most straightforward interaction, AI chats do fail for several reasons. Largely technical, things like supporting APIs that are not feeding data back to the bot in a timely manner or problems with the underlying services that are out of your control. In these instances, you want to enable your bot to “fail gracefully” and send a message to the customer that they are experiencing a problem and redirect the customer to another channel to receive support.
The thing you don’t want them to do is to go rounds with the bot saying, “I’m sorry, I don’t understand that” or “I’m sorry, I cannot help you right now.” In customer service bot programs that Neal Analytics has established, we try to come up with a graceful failure message or messages that are contextually appropriate to the situation experienced by the customer and provide them with another avenue for help. This could be a message like, “I’m sorry, I am experiencing a technical issue at the moment and cannot complete our conversation. Please contact our live customer service line at 1-800-555-5555 if you need immediate assistance.” Typically, you want to limit your customer interaction to three misses in a row then allow the transaction to gracefully fail.
Another area to anticipate the need for a graceful failure is when a user uses offensive language or curse words. Just as you don’t want your customers mistreating your live customer service representatives, you don’t want them to abuse your bot. The Microsoft LUIS Service allows you to identify certain words and tag them as conversation enders. In 1972, comedian George Carlin performed a monologue called “Seven Words You Can Never Say on Television.” This is a good place to start your list.
But seriously, ending a chat session at the sign of customer frustration indicated by a curse word is the right thing to do. It saves the customer from further frustration and again allows you to point them in a direction where they may be better served. A sample message in this instance would be, “I’m sorry, it seems I am not able to help you, if you would like further assistance, please contact our Customer Service Team at 1-800-555-5555.”
Including an AI chatbot in your customer service organization is a powerful tool that will bring consistency and broad availability to your service offering. Your customers will like the fact that they can self-serve anytime they want and on various communication channels.
The trick to bringing an AI chatbot to market is to do it so that your customers will embrace it as the convenient tool that it is. By presenting your AI chatbot as a bot, keeping the transaction simple, being prescriptive in your dialogue, and allowing the bot to fail gracefully when things aren’t working out, will give your customers the best experience possible.
Neal Analytics has deep expertise in creating user-friendly chatbots by leveraging tools like Cognitive Services: LUIS and Bot Framework, and Natural Language Processing (NLP). We have experience building chatbots for various industries like pharmaceutical, transportation, utilities, and others.
This blog has been updated and was originally published 1/18/2021.