5 Common Chatbot Fails and How to Prevent Them
“Sorry, I don’t understand.” “Oops, didn’t quite get that.” “Beep boop, can’t compute. Try again?”
If you’ve never received a chatbot error message, consider yourself lucky. It seems like for every one chatbot that works, there are five others that fail, further scarring its fragile reputation.
What’s going wrong? Is the bot at fault, or humans? Spoiler alert: It’s humans.
Here are the most common chatbot fails and how you can avoid them.
Conversation dead ends
Many of the chatbots used by businesses today are rule-based (vs. more advanced AI chatbots ), which means they make decisions and fulfill customer needs according to a series of defined paths. It’s ideal for chatbots that take care of simple tasks, but sometimes businesses keep its voice and decision tree structure simple too.
A little too simple.
If you neglect to inject any ounce of personality or program the bot to handle unpredictable commands, users are likely to get bored.
Simple programming can also hurt your chatbot’s performance. You can’t expect all users to follow your ideal conversation flow. Complicated questions are excusable, but if your chatbot is misunderstanding variations of basic commands and phrases, it’s embarrassing.
Solution: Brainstorm multiple conversation scenarios. An easy way to do this is by reading through chat transcripts or conducting qualitative interviews to learn how your customers talk, what terms they use, and so on.
Depending on the chatbot provider you use, this will help you create a comprehensive decision tree or conversation paths. It’s unnecessary to program your chatbot to recognize an extensive word bank, but creatively diverting off-topic requests can keep conversations from hitting a frustrating dead end.
If this amount of coding and programming would cost you more time and resources than you have, then consider using a no-code AI solution. With Userlike’s AI Automation Hub, for example, you just need to know how to copy and paste.rite conversations with multiple path options.
For extra help with perfecting your chatbot script, check out our post “6 steps for creating a smooth chatbot conversation flow.”
Do you know what’s more annoying than seeing your friend typing a response to you for a long time only to receive an “OK, cool”? A chatbot telling you their life story in one unprompted message.
Instead of striking a casual conversation with potential customers, these overzealous bots make elevator pitches with a hint of desperation.
When buttons or emoji get involved, it looks like straight up spam.
Buttons, text and links are useful for engaging customers and steering the conversation, but within reason. If a chatbot is overwhelming customers with floods of text, then they’ll be more likely to request an agent or abandon your site.
Solution: Create chatbot messages that follow the same etiquette used in live chat conversations. For example, break large text up into smaller messages and give the user time to respond before continuing.
You can also provide the user with commands and buttons to use at helpful times throughout the conversation. Commands should let the user switch topics or learn more specific details about something.
Buttons can also help speed up certain processes. For example, if a potential customer wants to schedule a consultation, the chatbot can ask for scheduling details using buttons for the date, time or desired meeting duration.
Commands and buttons also make it easier to avoid dead ends. You limit the risk of misunderstandings and put the user in control.
The Ultimate Guide to Chatbots in Business
Learn how chatbots work, what they can do, how to build one – and whether they will end up stealing your job.Read more
Under or overdeveloped persona
Chatbots are essentially digital robots, but that doesn’t mean they have to sound like a robot too.
Sounding too cold or stale can misrepresent your brand. If you hook customers with website copy that has a certain lingo and vibe, but your chatbot sounds like it’s straight out of the box, then you’re only hurting your image.
However, the grass isn’t necessarily greener on the other side. Talkative, overly humorous chatbots can have the same adverse effect.
According to our chatbot perceptions survey, customers want chatbots to crack jokes and show emotion. They say that, but it doesn’t mean they’ll actually enjoy it. Despite humor being universal, it’s also culturally specific. A joke may be a hit with users in Canada, but fly over heads in China.
A dad joke and pun here and there won’t hurt anybody though, and sometimes businesses get humor right.
Poncho, the dearly departed weather chatbot, was, in my humble opinion, a uniquely funny cat. But its fatal flaw was its tendency to lean a little too heavily on its humor.
This overuse of humor makes it take longer for a customer to receive answers or resolve requests. Not so funny anymore.
Solution: For chatbots that lack an identity, try matching its persona to the way your agents or target customers speak. Collect phrases and words your agents and customers most often use and try sprinkling them into the script. The more empathetic and personable it can be, the better.
When writing the script, it helps to create a profile for your chatbot. List its likes, dislikes, goals and problems to help guide you during the creative writing process.
Most importantly, make sure everything is grammatically correct and there are no spelling mistakes. Chatbots are meant to be intelligent after all.
If a funny chatbot is perfect for your brand, then be strategic with your humor. Chatbots are often utilitarian first, a “buddy” second. We recommend using humor in the beginning of a conversation with your bot to keep users engaged, then use jokes and quips sparingly once the user starts asking specific questions.
Above all, your chatbot’s conversation flow should be your first priority. Create a script that mimics human speech and then use jokes and cleverness to help sell its persona.
Doesn’t add value
Back in 2016, Business Insider reported that 80% of businesses want a chatbot by 2020. To be fair, this is based on 800 responses from business leaders in France, the Netherlands, South Africa and the UK. Not exactly “all businesses,” but it definitely foreshadowed the years to come.
Today, there are more than 300,000 chatbots on Facebook Messenger alone. But when so many easy-to-implement bots are hitting the internet, you’re bound to get a lot of duds.
A common fail among these bots: lacking a purpose, strategy and goals. Chatbots are attractive because they’re capable of taking on small, mundane tasks that once plagued employees. Not to mention their curb appeal.
But that modern, high-tech image of having a digital assistant is what businesses seemed to focus on too much. These chatbots may have a role, but it doesn’t perform well or add value.
For example, PayPal’s Virtual Agent makes it clear that it’s still learning, but offers to send additional resources if it can’t help.
When I told it that I need more help, it told me to contact customer support during business hours. It didn’t offer to conduct a second search, forward my message to an agent or give me an email or number to call. What’s the point?
Solution: First, determine whether or not you even need a chatbot. If it’s sure to tackle some pain points your agents, customers or company face, then clearly define its role and the actions it can realistically take.
For example, if you need to build a FAQ chatbot — or interactive FAQ page connected to your knowledge base, — make sure it’s well-versed on the questions customers ask often. PayPal has a whole page dedicated to email scams, yet its chatbot failed to recognize the keyword.
If it has one job, make sure it does it well. Regularly update your chatbot knowledge base so it can identify and handle all new issues and requests. And make sure you’re using a chatbot builder fit for your needs. Userlike’s AI Automation Hub makes it easy for anyone at your company to build a bot — really, anyone. You get all the most attractive AI features on your website without looking at a single line of code.
We highly recommend having a chatbot connected to your live chat solution so all unresolved inquiries can easily be forwarded to an available agent. This shows your customers that their questions are important and their time was not wasted.
This is the biggest chatbot fail of them all. When scripts break and programs go awry, chatbots lose their cool.
New chatbots are prone to mistakes in their early stages, but at the dismay of customers. You risk ruining your brand image and frustrating customers if you deploy a bot that’s not ready for the real world yet.
For high-stress situations, like fraudulent charges on your bank account or canceled flights, a malfunctioning chatbot is the last thing a customer wants to deal with.
Solution: Test, test and test your bot again before putting it online. Ask people outside of your team or a few customers to talk to your bot and give feedback.
When your chatbot malfunctions, lighten the mood by having your chatbot acknowledge its shortcomings ( “Sorry about that, brain fart. Let’s try this again” ) and provide options for moving forward, such as:
- Agent handover
- “Start over” button
- Menu options like “Ask a question” or “Report a problem”
Sometimes customers ask bots silly questions to test its conversation scope. If you’re using a rule-based conversational structure for your chatbot, create a rule that helps the chatbot handle these types of diversions.
This could be keyword commands the customer can use, the option to forward to an agent or button options like the Kasisto KAI bot uses below:
If your chatbot is prepared to handle a few oddities here and there, it’s less likely to break down and ruin the exchange.
If your chatbot fails consistently, try something new
Sometimes chatbots aren’t the answer. Some of the biggest chatbot fails the internet has seen ( Microsoft’s Tay anyone? ) came from bots that had a very narrow purpose or function.
There are many AI options on the market besides chatbots that help serve your customers. Make a list of the issues your service team currently faces and research software designed to help.
If you need a FAQ page customers will actually use, or help cutting down tickets, consider an all-in-one automation solution. Userlike’s AI Automation Hub offers three modules connected to a central knowledge base: an AI chatbot, Smart FAQ and Contact Form Suggestions.
Using a knowledge base and our self-learning AI will keep your chatbot from being a digital dunce.
And if your chatbot just doesn’t have the right answer, perhaps to a complicated personal question, you have your agents as fallback with our live chat software. Get a feel for Userlike with our free 14-day trial and reach out to us in the chat on our website if you’d like to learn more!