5 Pillars of a best-in-class chatbot
A 2020 review
Given the exponentially increasing investment of large organisations into new technology, companies feel like they need to give themselves a competitive edge — in 2020, where digital workforces are quickly becoming the new standard, chatbots have named the “tech du jour”.
How do you choose one? I don’t want to invest in something that’s not going to give me the best results.
There’s been a huge boom in tech startups creating their own platform to build a digital assistant — the choices are seemingly endless. Fact of the matter is most of these platforms offer comparable and equally impressive features like Natural Language Processing, flashy interfaces, social integration and AI.
The difference is how you write for it.
5 principles for great chatbot design.
1. Natural Cadence
When you ask someone a question, they might take a minute to digest what you’ve said, formulate an answer, and then write their response. This is a natural period of delay that we as humans expect when communicating to each other and to machines.
Artificial delays are a common trope of human computer interaction in order to bridge the gap between mental models and the digital systems we build. The technical term for this is called “Benevolent Deception” People expect a computer to have to process something, much like their brain does, however, instantaneous processing feels jarring as it’s perceived as too quick — the response seems canned.
The solution to this is to add a minimum of 0.1s delay to your responses.
2. Colloquial & Conversational
People understand that they’re talking to a chatbot, we don’t need to rely on overly artificial conversation to remind them of this — in fact — using conversational tone and language helps people feel more at ease when speaking to your bot, and will keep them more engaged.
We don’t speak the way we write. The experience of speaking to a chatbot should be no different than talking to a customer care representative.
Choreography of a conversation
Let’s break down a typical conversation to see how you can add nuance to your chatbot conversations.
Conversations are like a dance, that involve many moving parts, broken into distinct segments.
Greeting
Used to say hello or start a conversation. Formality is dependent on relationship (how familiar you are).
Asking
For engaging or seeking information. Helps keep the conversation going
Informing
Giving information that is either requested or pertinent to the conversation
Checking
Testing ones understanding Restating details for information and clarity
Error
When you don’t understand, or fail to fill a request
Apologising
Politely acknowledging your shortcomings. Should be brief and act as a bridge for alternate solutions.
Suggesting
Presenting relevant actions or alternatives
Conclusion
A clear end to the conversation
Keep things light and graphical
People make jokes, they make light of a situation, they emote, smile, send gifs, images, emoji or laugh to express how they’re feeling. Your chatbot should behave in the same way you would if you were having a conversation.
3. Reduce friction
There are 3 principles that relate to reducing friction:
If it’s easier to do something on the site than it is to do it via the chatbot, rethink your design
Define a user problem
What problem are we solving with the chatbot? Your chatbot needs to serve a real human need — too many organisations add a chatbot because the competition has one rather than because there is a real user need for one.
Added Value
What value are we adding to the customer experience? What other ways can this user need be solved? Are they easier/simpler than your chatbot? Does your chatbot replace those methods, does it augment them?
Communication
How are we communicating the value of this chatbot? Great! You’ve built a shiny new thing — but you need to show people what it can be used for, and why they must use it. One of the biggest barriers to chatbot use is visibility of services, if I as a customer do not understand how to engage, or what functions this bot serves, I am less likely to interact with it.
4. Efficient and Useful UI
With the advancements in natural language processing, bots are becoming increasingly able to understand intent through machine learning than ever before — as a result of this, more and more chatbots accept free text entry — the ability to ask any question you like in your own voice. This however, can lead to choice paralysis for the user, and again, relates back to what we mentioned about transparency. “What can I ask?” Adding suggested questions and follow up links shows users what your chatbot can do, and also lowers the entry barrier for new users to begin interacting.
Customers should be able to engage with a chatbot in any way they feel comfortable. This could be via prewritten messages, free text entry or carousels.
On the flip side of this, is providing too much, or an abundance of information in return. Again, imagine you’re typing to a friend, and they start sending back paragraphs in response, or lots of short messages in quick succession — it feels overwhelming, or spammy.
Responses should be brief. Long messages are overwhelming and might be an indication that a particular journey doesn’t belong on a messaging platform.
It’s important to not only break up your messages into digestible chunks, but to cut back on the amount of information displayed in the first place. This could mean deferring in some cases to external web content, or reducing your response to only the essential information.
5. Deep Understanding
Questions generally aren’t broad. They’ll have subtle nuances that indicate they’re asking about a particular aspect of a topic. It’s important when creating chatbot intents to break them down into sub-topics and help direct your user to the right one. There’s nothing more frustrating than asking a question and getting a generic response back.
The ChatBot should follow up and continue to ask questions in order to narrow it’s answer to a single response. Try to avoid general statements or multiple topics in one message.
A good way to make sure you’ve got all topics covered is to speak to people, and ask them what they know about a topic. See what types of questions they ask, and what they do and don’t know. If you’re writing for a topic you understand very intimately, this will give you a fresh perspective from someone with a way of thinking that closer matches your audience.
Just like with any piece of work, good design is not a set and forget process. You’ll need to test, readapt and redeploy your chatbot on a regular basis so it can continue to improve over time.
“Never stop testing, and your work will never stop improving” — David Ogilvy
