How to deal with errors like a bot
Tips and tricks for dealing with mistakes
Iām sure Iām not the only one who absolutely loves that little backspace button on my keyboard. I donāt know how often Iāve pressed it by now, but itās definitely A LOT.
Sadly, there is no backspace button in real life.
You can prepare oh so well, but in the end, weāre only human. We do weird things and make mistakes all the time. I remember sitting down at a tiny restaurant somewhere in the South of Thailand and looking at the menu, not understanding a single word of what I was trying to read. So instead, I just pointed at something and ended up eating a very dubious-looking dish that Iām pretty sure had chicken feet in it. The lady who cooked and served it kept staring at me, so I had no other choice than to finish the entire plate, thank her, and continue my journey feeling a little bit nauseous.
What Iām trying to say is, mistakes happen all the time, itās how you handle them that matters. The same goes for chatbots. No matter how well youāve trained them, bots deal with people and people are very unpredictable. We express ourselves in a bazillion different ways and say the weirdest things (especially to bots) so itās kinda impossible for your bot not to make any mistakes.
So how can you deal with these errors gracefully and efficiently? By making sure you have a few good error messages built into your conversation. Not sure how to do that? Iām here to help! Here are my best tips and tricks on creating kick-ass error messages:
1 ā Keep calm and clarify the mistake
The first step to writing a good error message is to first acknowledge and then clarify the actual error. Just own up to the mistake. If your bot is not able to understand a request, let the user know, but please donāt act as if itās the WORST THING THAT EVER HAPPENED. Overly apologetic bots are not helpful, theyāre just reallyā¦ annoying. Also, most users understand by now that bots are virtual things that mess up every now and then. That being said, they do expect you to clarify the mistake and provide them with a way forward.
Which brings us to the next tip.
2 ā Provide a way back to the āmain menuā
When using more traditional graphical user interface (GUI), it has become second nature for us to look for the home button whenever we want to start over. Command+r please. But when it comes to conversational interfaces, starting over can be a bit more tricky, though it shouldnāt be impossible.
Letās say youāre building a chatbot for a product platform that can help its users do the following tasks:
- Explain pricing models
- Show product features
- Book a platform demo
If I ask this bot to āgive me the number of the CEO,ā the NLP probably wonāt recognize that intent ā at least, I hope not. This use case it isnāt part of the original bot scope, so the bot wonāt understand the question and will need to fall back on an error message:
Sorry, I canāt help you with that, but I can tell you all about pricing and product features, or book a platform demo to get you started straight away!
Sounds good, right? With an error message like this, the bot not only shows empathy and understanding, but it also provides the user with a way to continue the conversation.
Another way to guide the user back to the āmain menuā is by adding a āstart overā intent with expressions like āhelpā and āstart overā that trigger the āmain menuā or introduction message.
3 ā Write various error messages
One of the best practices of conversation design is to create different variations of bot messages: various greetings, various responses to chatter and yes, you guessed it, various error messages! No one wants to hear the same message over and over again ā it causes a bad user experience and feels way too robotic ā two of a botās biggest sins. So make sure to create various error messages to keep the experience conversational:
User: I need a massage.
Bot: Sorry, I didnāt get that. Can you rephrase?
User: I just want you to massage me!
Bot: Sorry, I still donāt understand. Try one more time?
User: MASSAGE, NOW
Bot: Sorry, I canāt understand. Would you like me to connect you with customer service instead?
4 ā Make sure you have a back-up plan
If your bot is really stuck and canāt understand its user, make sure it can reroute them to another line of support. This could be an FAQ page on your website, a phone number to reach customer support, or even an email address where they can send their question to.
5 ā Add some humor (but tread carefully)
I love a funny bot, I really do, but it all depends on what Iām trying to achieve with it and how important those tasks are. For example, if Iām trying to transfer my house rent using a banking bot and it keeps messing up, I wonāt be in the mood for a joke. But if Iām using a retail bot to browse for new clothing and it canāt seem to find the āperfect white shirtā, I donāt mind a little joke to lighten things up. As my CEO likes to say: āwhen in doubt, leave it outā. Iād like to add: āBut when appropriate, go for it!ā
6ā Make your users feel heard
If your bot canāt help its user, at least make sure they feel heard. Ask for feedback! I always include a little feedback flow into every bot I build, not only is this super-duper handy for collecting valuable feedback on the botās performance, but it also shows the user that, yes, we do care about your opinion and we do want to hear what you think. People really appreciate it if you give them a chance to share some feedback.
Right, I think thatās about it! If you have any questions, feel free to get in touch. And if not, carry on and happy bot-building!
Hi there, Iām Tess, language lover and taco enthusiast! Currently Conversation Design Lead at Chatlayer by Sinch, I like to write about the things I learn and see when building multimodal bots.
Interested in knowing more about voice and chatbots? Want to swap ideas or talk conversation design? Reach out to me on Twitter!
