Building a chatbot is a lot of fun. It’s a cool way to bring a touch of modernity into websites, and also a valid strategy to improve online businesses.
However, the reality is that lots of chatbots are pretty awful, and can worsen customer experience in a number of ways.
Companies started noticing this too.
Looking at the chatbot landscape, you can tell that the “chatbot fever” we saw a few years ago has toned down. High-profile incidents involving chatbots built by the likes Microsoft and Tencent didn’t help either.
So, what makes a good chatbot? Or better yet, why do chatbots fail?
In this article, we explore the top reasons that prevent chatbots from being useful. We are confident that considering these will help you build better chatbots, or else provide you with a broader understanding of how to improve the ones you already built.
Bad chatbots: Examples and solutions
Chatbot fail #1: Building a chatbot for the sake of it
Chatbots can be cool, but that doesn’t mean every single website needs one.
How many times have you run into chatbots whose sole function is to list out navigation items and share an email address in case you have a request?
In most cases these are pointless use cases, as users can just as well browse your website menu. Moreover, if users are having trouble finding stuff around your site, the problem is likely not the lack of a chatbot, but a dysfunctional website.
So instead of jumping the gun, it’s wise to ask yourself whether your business needs a chatbot or not.
Would a chatbot provide value to your customers?
Is it bringing something that your website or store is currently missing?
If so, what is it?
If not, why create a chatbot at all?
Answering these questions will help you shape your chatbot strategy, define your goals and find out what to measure.
Chatbot fail #2: Overestimating chatbot abilities
Artificial Intelligence (AI) and Natural Language Processing (NLP) are constantly evolving. You can already witness shockingly natural conversations between AIs.
However, commercial chatbots are nowhere near this level yet. Expecting your chatbot to handle full-blown conversations will inevitably result in frustrated customers.
The perfect example of this is Poncho, the now-defunct Facebook weather bot.
Designed to provide weather forecasts in a fun, conversational way, Poncho disappointed users with its off-point, hilarious exchanges.
The lesson here is simple: Unless you are, or happen to employ a cutting-edge AI engineer, be realistic about the technical limitations of chatbots.
Most chatbots are unable to understand spontaneous interactions beyond picking up a few keywords.
To avoid frustrating customers, you can leverage other ways to create conversation-like interactions, including buttons and polls for example.
Providing clear instructions and a well-defined conversation structure will allow customers to embrace the flow of the conversation, and prevent trolls from asking irrelevant questions (more on this later).
Hubspot is one of the companies that got this right. In the example below, you can see their a bot driving the interactions forward by providing clear options the user can choose from.
Chatbot fail #3: Giving a robotic, automated feel to the conversation
No one enjoys talking to a machine. Think of the last time you had to deal with an automated phone system, listing out options and guiding you menu after menu with a rigid, pre-programmed voice.
This is not what you want for your chatbot, right?
To solve this, let your creative juices flow. Give your chatbot a personality to make the conversation feel more natural, and be conscious about the choices and consequences.
You can achieve this with a few steps. Start by giving your chatbot a name and a picture, then build a persona based on your business.
A bubbly virtual assistant could be a great fit for an online candy store, while a polite digital butler might be a great choice for a hotel chain.
The point of “chatbot personalities” is making the conversation feel natural. Adding small pauses between messages is a great way to mimic human behaviour, as if your chatbot was typing. Such skeuomorphisms are a great way to engage users, creating the "real conversation" feel.
Take the cosmetics chain Sephora, for example.
It uses a chatbot to recreate its in-store experience on its website by using a friendly tone and asking questions a sales associate would likely ask. Their chatbot helps move users down the sales funnel, from general questions to purchase.
Chatbot fail #4: Pretending your bot is human
Building chatbots is like walking on a thin line: You don’t want them to sound robotic, but you can’t pretend they’ll be able to pass the Turing test either.
The current state of chatbots doesn't allow for true organic conversations, though pretending your chatbot is human will backfire sooner than later.
The advice here is to give your chatbot a personality and charm, but be upfront about its limitations.
Transparency will help you manage customer expectations, and prevent them from getting frustrated.
Chatbot fail #5: Trying to automate everything in the conversation
As much as we love automation, there are limitations that need to be considered.
At one point or another, a degree of human involvement will be needed. Then what? The last thing you want is for users to feel stuck.
The key to this problem is to give the user the necessary options when a human needs to be involved in the conversation.
Do you want your chatbot to handle the bulk of customer support, serving as a filter for your team? That's totally fine, and a great strategy. However, don't chase automation at the expense of your customer experience.
No one likes to be stuck with a robot.
Chatbot fail #6: Overwhelming users
Chatbots are a great way to provide a lot of options to your customers, and to use as a distribution channel. From your latest ebook to your upcoming webinar, passing through the ability to contact support, and let’s not forget about that product launch you're planning...
It can become overwhelming for customers.
There’s a good rule of thumb to deal with this situation.
If the chatbot is sending answers and they don't fit in the messenger window, it’s going over the top. It's not a conversation anymore, but an information dump.
The advice here is to keep the conversation within certain extension limits while focusing on what matters the most.
Also, make sure the bot breaks up the information into smaller chunks to make it digestible. Sometimes, that means turning a long paragraph into 3 messages and including a few buttons:
Chatbot fail #7: Creating a troll-friendly environment
There will always be people trying to exploit any loophole they can find.
If you decide to use open text fields in your chatbot, you are bound to experience some trolling. As a business, this can land you in murky waters, and set you up for some shameful, viral screenshots.
The most effective strategies to deal with chatbot trolls are removing open text fields and using bullet-proof copy.
Copy should be neutral, and feature multiple bumpers to prevent conversations from going off the rails.
Chatbot fail #8. Putting users in a loop
You might think that once you get your great chatbot diagram the game is over.
The sad truth is that customers will take your chatbot to places you’ve never imagined. And here’s where loops usually appear.
A stuck user is an angry user, and because there’s no human involved, it all stops there.
To address this, provide alternative texts for the same node and make sure to track whether one specific message is sent disproportionately. This could mean your decision tree has an issue.
In addition, prepare your chatbot to offer a “way out” when certain parameters are met (these can be keywords, repetition, and other cues).
Chatbot fail #9: Stopping the development cycle after launch
Kudos to you if you manage to create a comprehensive conversation flow and set it live, but the work doesn’t end here.
Like most software, chatbots are better when updated, improved, and revised.
Things will come up as customers interact with your chatbot, and you want to make sure you improve the experience as new data starts pouring in.
Sharing a survey with users at the end of every conversation is a great way to passively collect user feedback and look at the resulting data to assess satisfaction.
Final thoughts: Strive for purpose, not for attention
Chatbots can drive tons of value when done right, but they require a bit of work and patience.
As we’ve seen throughout this article, it all comes down to the very first step: Why do you want a chatbot and what purpose will it serve?
This question will be your lifeline throughout the chatbot creation process. It should guide the rest of your efforts, from crafting the copy to creating the flow that will help your customers get what they need.