This month, we’ve been looking at chatbots for digital customer service. We started off with our summary of the excellent Dashbot panel on building enterprise customer service chatbots. Next, we looked at chatbot trends and facts to know for 2018 and chatbot customer service success stories. Today, we’ll close off our series on chatbots by reviewing some buzzword-free customer service chatbot advice.
Why? Because there’s an awful lot of advice about customer service chatbots out there, and a lot of it focuses on buzzword-driven SEO-friendly copy that doesn’t provide real, actionable advice to companies looking to get started on their own customer service chatbot journey.
1. Chatbots and digital assistants are not interchangeable
Perhaps one of the areas where buzzwords create the most confusion is in talking about technologies behind virtual assistants like Siri and Alexa. With the increasing popularity of smart speakers like the Amazon Echo and Google Home, there is often a tendency to conflate chatbots and digital assistants. But that overlooks the fact not only is the technology different, but so are the applications.
Developing something with the capabilities of a digital voice assistant takes a lot more time and effort than developing a chatbot, because chatbots don’t need to be able to do everything that digital voice assistants do. Customer service chatbots don’t need to support open-ended conversation, and the extra time required to develop recognition of intents acts as a barrier to being able to execute on simple menu-based flows that will alleviate a large volume of common information requests that human agents find tedious to handle.
2. Not all chatbots need AI or machine learning
Some of the other most-used buzzwords that get thrown around in relation to chatbots are “AI” and “machine learning”, but this also does a disservice to companies who are looking to get started on their chatbot journey. Not every company needs a chatbot with AI or machine learning capability! For companies with high complexity of product and service offerings, such as financial institutions, insurance companies, or media services companies, a bot that learns from interaction with users can be beneficial, because machine learning can help highlight issues that might otherwise get lost in a complex customer service landscape. But for companies with a low-to-moderate complexity of products and services, who regularly field a high volume of routine, easy-to-answer informational requests, machine learning won’t deliver any noticeable benefits.
When considering whether your bot would benefit from machine learning capabilities, it’s important to remember that developing a bot that will learn from user interaction over time is a much larger task than developing a simple menu-based bot that can guide users through a process of getting information to the most commonly-asked questions without needing to use machine learning to assist customers. And implementing machine learning will greatly increase the time needed to develop a bot.
Learn from the case of Westjet’s travel assistant bot, Juliet. There are a lot of simple bot use-cases that don’t require the extra development time to develop something complicated, so start with those and build out from there. As people interact with the bot, you can expand the supported use cases over time. And, down the line, should you decide that machine learning would be an asset, you can always upgrade the bot’s capabilities. But when you’re just starting out, small and simple is might be best.
3. Don’t try to mimic natural conversation
NLP (Natural Language Processing) is another popular buzzword that gets thrown around whenever the topic of chatbots comes up, but the truth is that if you’ve never deployed a chatbot before, you’re probably better off avoiding a purely conversational bot for several reasons.
First, while Natural Language Processing has improved greatly in the last few decades, it still is far from reliable. Even well-designed bots that use NLP still struggle in many cases to understand common user intents – which can end up increasing customer frustration rather than alleviating it. And, again, conversational bots also require more effort to develop. Extra time has to be taken in defining and testing common user intents, because even your most commonly asked questions can be phrased in many different ways.
A good example of this is Margot – the wine recommendation chatbot developed by UK grocery store chain Lidl. When customers ask Margot for wine recommendations filtered by price, it had to be able to recognize intents for “all of the following: ‘£5’, ‘5£’, ‘5 quid’, ‘a fiver’, or even ‘around five-ish’.”
Second, it’s important to consider that chatbots for customer service are still quite uncommon, and there are many potential users who aren’t outright opposed to using chatbots to answer customer service questions, but who will also need guidance to feel comfortable with the process. A purely open-ended conversational chatbot can leave customers unsure if they are interacting with the bot in the correct way, whereas a bot that combines conversational flows and menu-based flows can help customers feel more comfortable that they are getting the correct information to help them with their issue or decision. Remember that the ultimate goal of a chatbot is to make your customer experience easier, and design with that goal in mind. Your chatbot needs to either guide users through a decision-making process or guide them toward connecting with a human agent who can assist them.
4. DO give your chatbot a personality
People don’t want to have emotionless interactions with brands; customer decisions are driven by emotions, so it’s important that customer service chatbots convey emotion and personality – which is going to require involving teams outside of development to ensure the tone of your chatbot is consistent with your brand. As mentioned in a previous post (link):
It’s also important to ensure that writing of bot scripts not be left to developers, because a customer service chatbot will necessarily need to be friendly and consistent with the personality of your brand. Chatbot scripts need to be written by people who can write natural-sounding language that conveys personality while still representing your brand professionally. For many organizations, this means having experienced marketers and copywriters handling bot scripts, but some organizations are looking further afield to professionals like screenwriters!
5. Balance wordy correct answers with brevity
Customer service chatbots don’t need to have complex or highly technical interactions with customers; they simply need to make resolving inquiries easier and more pleasant than it currently is. An important part of that is respecting customer communication preferences for digital messaging.
If a customer reaches out to your chatbot for an answer to a customer service question, it is highly unlikely that they are looking for a highly detailed or complex answer to their question; if they required a highly detailed answer, it is most likely that they would reach out to a human agent through phone or email. Further, when interacting over digital messaging, customers will engage from wordy “text walls” – so keep your answers concise and to-the-point. If it turns out that a customer does desire more complexity of answers, they can always ask to be transferred to a human.
6. Make chatbots optional, and always be able to hand over to a human
An important part of designing chatbots for customer service is making it possible for customers to be able to easily contact a human for assistance at any point in the chatbot interaction. While it’s true that one of your primary goals in deploying a customer service chatbot will be to divert routine customer service traffic from your human agents, it’s important to remember that it’s important to delight your customers rather than annoy them. While many customers will be open and receptive to using chatbots for customer service, there are customers who won’t be and who aren’t going to change their minds. So make sure that your chatbot doesn’t become an obstacle that annoyed users have to bypass to talk to a human agent – rather, make your customer service chatbot an opt-in process for those who want the convenience of instantaneous service, and let customers who would rather connect with you over a different channel do so.
Additionally, there will be a portion of your customers who are open to experimenting with the bot but who will find it difficult to complete the interaction in the way that they need. Enabling them to transfer smoothly and easily to a human agent will be important, as will ensuring that the human agent who accepts the bot handoff has access to that customer’s bot interaction history. This way, the agent will be able help the customer without asking them to repeat information, which would frustrate your customer.
7. Always monitor and be prepared to adjust bot output!
A chatbot isn’t something that you can simply set and forget! It’s important to always have humans monitoring chatbot interactions and output to ensure that the interactions that the bot is having are consistent with your strategy and aren’t annoying your customers. This is especially important if your bot employs machine learning, because the results of machine learning algorithms can be unpredictable when deployed with users on the internet.
The best cautionary tale is the case of the 2016 release of Microsoft’s now-infamous chatbot Tay – which was designed to mimic the conversational style of a human teenager on Twitter. Tay was built using machine-learning algorithms, and was supposed to learn to converse more naturally through interaction with humans online. However, in what ended up being a PR disaster for Microsoft, Tay was deactivated after only 16 hours due to coordinated effort by online trolls. Trolls sent Tay thousands and thousands of racist, misogynist, and pro-Nazi messages until Tay was sending out shockingly racist tweets:
Granted, what happened with Tay was a pretty extreme case. But it’s probably best to ensure that someone in your organization is responsible for regularly monitoring bot interactions and output for potential problems, so that they can be addressed.
That wraps up our series on chatbots. We hope you’ve found it interesting and informative! If you’d like to talk with one of our digital engagement experts about how you can get started on your digital customer service journey, be sure to contact us. Or book a demo to see ITC in action.