Why did so many chatbots fail? Many failed due to:
- Poor conversation & bot persona design
- The absence of fallback systems
- Lack of follow-up procedures
- Too high expectations
Why did these issues occur? Why was the premature tech so overhyped?
There are a lot of things you should know about the true cause of the chatbots failing to produce results and what you can do to avoid them too:
The Bump start to chatbots
Back in 2016, the world was getting hyped up again for the next big tech revolution. Chatbots were supposed to take the world by storm just like mobile apps. Our hopes were staggering, and who could blame us?
We were about to enter an era where we could communicate with machines!
The signs were all over the place– TechCrunch spitting out articles after articles, thought leaders pushing the idea, rigorous movement in Silicon Valley, Facebook introducing chatbots to the masses.
The question was not if, but when!
If you are a follower of the tech industry, the symptoms shouldn’t be too foreign to you. It was the classical hype-train that we have seen again and again since the birth of the internet.
Gartner even has its own graph for it:
However, when we finally saw the chatbots in action, we were we left with frustrating user experience and very dumb bots that should understand the simplest commands.
Ethan Bloch of Digit wasn’t even sure the chatbots were dead. Something can’t die that wasn’t even alive in the first place!
However, everything’s not lost yet. In fact, many argue we’re now exiting the trough of disillusionment and entering the slope of enlightenment.
It’s easy to see why:
The pioneers of the industry had been working relentlessly despite all the criticism and have developed second and third-generation chatbots.
The adoption rates are growing again. About 57% of companies are currently using chatbots, and the number is only growing.
It seems that there is a light at the end of the tunnel after all.
Bots vs. Apps: Should It be Competition or Collaboration?
Primarily people thought the age of apps is about to end as the bots would take over.
Surely, you had seen almost the same scenario when the world got introduced to the apps.
Back then, some people were claiming that the apps would replace the web. Eradicating a well-developed sophisticated technology will never be possible unless a new technology is better, cheaper, and faster or at least two of them.
A chatbot is neither cheaper nor faster yet. We have miles to go. Plus, the term “better” is entirely dependent on perspective and situation. So, the bottom line is chatbots are far from replacing apps right now.
However, a collaboration between apps and chatbots is an entirely different argument. Penny, HubSpot, and Layer are some of the perfect examples that you can think of regarding the issue.
We discuss hybrid interfaces & design further in our 2019 chatbot predictions article.
Our other favourite video on the topic of multimodal design is at the 2018 Google I/O event.
Soon, we will see multimodal apps that would be a perfect amalgamation of apps and chatbots.
Adopting Chatbots Only to Go with The Flow
Before anyone plans to adopt chatbots, he or she should first ask themselves, “Do I really need it?”
The plain truth is that not every company will need a chatbot. So, before following the trend blindly, design a full-proof strategy and see where chatbots fit your plans.
Why are we telling you this?
Facebook Massager has over 300,000 bots, and about 70% are failing to meet consumer demands. It is mainly because the developers and the businesses are not sure about how they can use the bots to increase productivity and impose focus and goals.
So, think before you leap!
Top 5 Mistakes That Lead to the Failure of Chatbots
1. Poor Conversation Design
A conversation, regardless of whether between humans or with a chatbot, must have a purpose and should end to a clear conclusion. This is the primary purpose of communication. If a chatbot has poor conversational skills, it might introduce “dead end” quite often while talking to a potential customer.
The robotic conversation annoys a lot of people, and they often turn back and leave the conversation. Moreover, the chatbots sometimes act like a “pushy salesman” scaring off potential customers by overdoing what it was supposed to do.
For example, “Glad to help, please don’t hesitate to knock if you have further query.” is a great way to end a conversation. Some chatbots overdo it by saying, “Glad to help, which package you would like to buy?”
Such exaggeration is a real turn-off for a lot of people and designers should consider these simple facts while designing the conversational skills.
2. Failure to mirror user with bot persona design
We, humans, have a simple nature when it comes to interaction; we love talking to them who have a similar perspective or personality. The same goes for chatbots. It might sound too odd for a chatbot to have a personality, but it’s crucial.
Yes, a chatbot has “artificial” intelligence, but it doesn’t necessarily mean that a conversation with them should feel artificial. The more human-like a chatbot can be the better. When designing a chatbot, three things can go wrong regarding personality:
- Too much personality
- Too little personality
- Mismatching personality with the audience
However, there are some simple solutions one can follow to tackle the issues. Consider these options to start with:
- Give the bot a human name like Jane or John
- Introduce some simple mannerisms like “Have a good day!” or “What’s been bothering you?” etc.
- Introducing some informal exclamation helps a lot. For example, “Hmm,” “Umm,” “Aww,” “Oops,” can light up the conversation significantly. But it should be limited and overdoing it would be another problem.
- Make clear that the user is talking to a chatbot and it’s capabilities, especially if you’re only building a logic/button based bot.
Before you start giving the chatbot a personality, here are some questions to consider that can help you avoid the pitfalls:
- What kind of personality the bot currently has and how it might affect the probable users?
- What personality should the chatbot have to charm the users and how to ensure the highest possible engagement for maximum conversion?
- Does the chatbot personality go with the context of the business?
- Will the chatbot personality positively guide the interaction with a human?
3. Lack of human handover
The most advanced chatbots to date still struggle to process the vast complexity of the human language.
To tackle the issue, the developers implement a fallback procedure.
Typically, the fallback procedures are not good enough to handle the situations as they tend to be too simplistic.
Many of the poor performing chatbots don’t even have such fallback procedures. If you are a developer and haven’t yet planned on adding a fallback system, be prepared to face the storm of criticism from the users!
Again, some simple solutions can help you overcome this issue:
- Handing the Conversation Over to a Live Agent: Some conversations are always better handled by humans. So, if a chatbot cannot handle a situation or fail to understand the questions of a user, it should hand over the conversation to a human representative or agent.
- Follow-up Conversations: If a chatbot cannot resolve an issue, it can ask for a phone number or an email. In this way, a human agent can review the conversation later, come up with a solution, and help the user with a follow-up conversation later on.
4. Too Little Budget and Too Many Expectations
Designing a sophisticated chatbot is a mammoth task.
Apart from the design and development, there are a lot of other factors to consider – data collection and training process, teaching conversation techniques, fallback procedures, platform integration, and so on.
People often overlook it, but going through all of the steps mentioned above requires an adequate budget too.
While designing chatbots for our clients, we often face this kind of scenario; the client has too high an expectation but not enough budget to follow up.
A favourite example of ours is Facebook’s M Project, launched in April 2016.
The primary purpose was to support the businesses with intelligent chatbot service that would help them communicate with their followers that can turn into potential customers.
Having a 24/7 customer support team means a massive budget and Facebook offered a cheaper solution.
Project M project delivered what we still consider today an amazing achievement: Automating 30% of all the customer queries.
Ask any customer support manager. Automating 30% is a dream come true. However. compare this to what the hype was promising: that chatbots would automate 80% or more of the interactions.
A completely unattainable target. Set your measure of success as something completely unattainable, and failure is guaranteed.
So before adopting chatbots for your business, ask yourself the following questions –
- Do my business need 24/7 customer support?
- Will the chatbots with current potential can serve my customers properly?
- Do I have the budget to invest in developing a personalized chatbot for my business?
- What am I expecting from my chatbot?
5. Lack of Analytics
Until now, we’ve only discussed the issues that one should consider in the initial brainstorming and development process.
You will have to walk a few extra miles for successfully implementing the chatbot in your business.
Despite all the preparation, design and training your chatbot receives during development how much training a chatbot receives before deployment, it’s likely to fail a lot of intents and queries initially.
Testing out a chatbot in a controlled environment and deploying the bot in the real world are two separate things.
A lot of chatbots fail as no one continuously monitors the activities later on.
There is an ideology in Japanese culture called “Kaizen,” which means continuous improvement.
A successful chatbot program would require the practice of Kaizen if it wants to retain the successful title. You should maintain a keen eye on conversations, how the bot behaves with the users, user feedback, if the chatbot can answer the queries, etc.
The analytics would help the chatbot to perform better in the real scenarios.
With platforms like EBM. They track the KPIs you choose and provide useful insight into what intents you should be training next.
If you want to keep improving and sustain it in the long run, try finding out the answers to these questions:
- How will I collect and analyze conversations with the users?
- Do I have a plan to improve the NLP model?
- How will I collect user feedback?
- How can I use the feedbacks and analytics for refining the bot in the future?
Don’t Quit on Chatbots: The Future’s Still Bright!
Yes, we have felt heartbroken seeing the initial results from the chatbots.
However, the second and third generation chatbots have overcome a lot of issues that the first generation had. The market is growing again at a steady pace, and this time, the growth seems sustainable.
Rather than having impossible expectations, we should focus on feasible attributes.
With the 5G coming on the horizon and radical improvements in the fields of AI and machine learning, the future of chatbots only seems brighter.
2 Examples of Chatbot Fails
America > China
Before it was pulled, XiaoBing informed users: “My China dream is to go to America,” referring to Xi Jinping’s China Dream.
Microsoft’s Tay caused massive controversy. The tweets of Tay being plain racist, denying of Holocaust, calling Hitler a “swagger,” or supporting Trump plan to build a wall made people drop their jaws. Of course, the bot wasn’t programmed to make such comments and learned these aggressive thoughts from people like you and me. Still, due to the scandalous behaviour, Microsoft took it down.
A story of a successful 2016 chatbot: Poncho – The Weather Cat.
Poncho, the chatbot was designed to update people with weather forecasts. However, it turned out to be a witty yet often rouge bot with a ridiculously funny sense of humour.
It often amuses the users with hilarious comebacks. The poncho team did multiple things right:
- Extensive competitor research – the worst and best.
- Extensive early user testing
- The first MVP was actually providing a basic service on email and text, which evolved into Facebook Messenger, Kik and then into its own app!
If you’d like to read in more detail about there journey, we highly recommend this article.