The conversation designer’s handbook — or how to design chatbots, Google Home actions and Alexa skills that work
For some time I’ve been meaning to put together a guide for writers and designers working on conversational interfaces based on my experiences designing bots over the last eighteen months.
While I haven’t quite found time to do that yet, I’ve managed to collate a list of what I believe are the best articles and resources on the subject.
Feedback and suggestions welcome.
Contents
Writing as a design discipline
The challenge of creating a great chat bot experience remains a storyteller’s domain. There will always be a market for writers who can jump from genre to genre. The next great written genre is interactive. It looks and feels like a conversation.
You need to do three things to be able to compress information, regardless of the domain: (a) First, you need to know the objective in terms of what it is you are trying to achieve; (b) Secondly, you need to understand the domain of the problem and the solution well enough; (c) You need to have the communication skill to be able to understand what is the minimum amount of information of the problem or solution that needs to be communicated in order to achieve the objective that you desire.
With the rise of voice interaction, now more than ever, our choice of words will influence how people perceive the customer experiences we design for them, because there are no accompanying visual cues to serve as a guide. Designers for the voice context must realize they’re relying 100% on what the user perceives the chosen words and phrases to mean — a notoriously squishy concept!
Designing for voice could be the catalyst that helps us return to the original goal of UX: treating people like people again.
Business process redesign and better training are important, but better use cases — those real-world tasks and interactions that determine everyday business outcomes — offer the biggest payoffs. Privileging smarter algorithms over thoughtful use cases is the most pernicious mistake I see in current enterprise AI initiatives. Something’s wrong when optimizing process technologies take precedence over how work actually gets done.
Beyond the buzz, what conversational UI shows us is the idea that everything in design should be considered a conversation. All elements in an experience are exchanges in a conversation amongst people, a brand, a business, a service, and, increasingly, a machine-powered intelligence. In this sense, conversation is the primary design material for designers — and its general purpose is to impact or inspire people’s behaviors and emotions.
Choosing the right use case for your bot
A bot is a type of user experience and a way to expose products, services, or a brand. The only way to make money out of bots, without having a service or a product, is to be a bot builder and have someone pay you to build that bot. End users to do not pay for bots, they pay for the services they expose.
Content chatbots are a big opportunity for brands to deliver their story to consumers. There is more clutter in all online and offline channels than ever before in history. And consumers are harder to reach than ever before. There is one lesson I think can be learned from the history of advertising: being first with good content in a new channel is priceless. Chatbots present this opportunity. Yet only if done right.
What is needed is a far more natural and conversational online shopping approach and experience. How chatbots bridge the gap between the stateless search engine query and a shopper’s actual intent is a challenge that has many of eBay engineers’ undivided attention. And for good reason. Crossing this chasm will amount to a scientific accomplishment equivalent to the Holy Grail of e-commerce.
Don’t just build a chair, build a chair so people can rest on it.
The last question to ask is — Is the cost for using your chatbot-powered product or service less than the perceived value. By cost I don’t mean just the monetary value but instead the effort, attention, time and (perhaps) money it will require of users?
There are two main reasons why anyone would use a chatbot:
Conversational: When an App can’t do it because multiple variable inputs are needed to solve the problem.
Simplicity: When a bot offers the most immediate and direct solution to a person’s problem.
Does your bot need a personality?
Bot personalities need to be built in reverse — from the user goal and ‘job’ backwards. Instead of a human deciding which career suits their persona, a bot’s job gets chosen first. Once you have established your bot’s goal, and background, you can determine a personality type, and traits that will guide your dialog development.
Marketers craft brand narratives, creative writers plot a storyline, and copywriters make micro-copy magical. Conversation designers are a hybrid of all three. Don’t underestimate the importance of words.
- Branding Bots Part 1: Crafting Voice and Tone For Your Chatbots and Voice Interfaces (Lauren Golembiewski)
The core values and mission of the company are the pillars of the brand’s voice. Voice should stay consistent and in alignment with the company’s values and mission. Bots are like an ambassador for the brand and they have to engage appropriately. The voice of a bot can be sarcastic, cheerful, or any other qualities you think reflect the core mission and values of your company.
It is clear that the utilisation of bots to complete routine tasks (such as banking, information-gathering, scheduling etc.) is a growing phenomenon that will continue to pervade people’s daily lives. Because people essentially learn through modelling and representation, we have a very real opportunity now to affect the society through the disruption or reinforcement of gender stereotypes by the way bots are designed and the ideologies they inevitably represent.
“Gender adds to persuasion. It also comes with a ton of cultural meaning,” said Jason Alan Snyder, chief technology officer at brand experience agency Momentum Worldwide. “We don’t need gender to humanize things. We’ve been talking to objects forever. Now they talk back. Gender assignment is something we need to be very considerate about. There’s great risk in amplifying negative things about society and moving them forward at scale with these technologies.”
Buttons vs Natural Language Processing (NLP)
One of the beauties of conversational interfaces is the ability to send in unstructured data — images, videos, button clicks, and more importantly a user’s own voice — their own words telling you what they want from your bot and what they think of your bot.
The right combination of free-form messaging versus buttons and quick replies can depend on the use case. Categories like travel, sports, and insurance tend to have higher engagement with the right mixture of buttons and quick replies versus free-form text.
Writing dialogue
Chatbots in B2B have their function. People visit such websites for a particular reason, because they want something. It’s like going to a restaurant or entering a bricks-and-mortar shop. Of course, sometimes people do it because they have nothing better to do or they just want to amuse themselves, but generally — there’s some purpose behind it; ordering food, buying a pair of shoes, or learning about prices. On the flip side, a waiter or a shop assistant also have their tasks and scripts to follow when talking to a client. A conversational website can work exactly the same way, and a chatbot’s role can be similar to a shop assistant or a waiter.
When designers work on websites or applications, we think about the visual hierarchy we should assign based on user goals and needs. If we do it wrong, users become distracted by other less relevant elements, and become lost on the journey to complete the task for which they first came to our website.
With a chatbot, the process is not different. What varies here is the set of UI elements used: text, cards, images, emojis and quick replies. Initially, these elements may seem a bit limited, but they are more than enough to make users want to chat with your bot.
When designing a product for the web or mobile world, we usually operate within certain types of interactions and user interface elements — text fields, forms, buttons, checkboxes, or switches, for example. In the chatbot world, the interface elements are different.
So, after you come up with lots of ideas about the magical things your bot might be able to do, it is time to get “down to earth” and balance your goals with the design constraints of the platform.
At their simplest, design principles are a list of strongly-held opinions that an entire team agrees on. They force clarity and reduce ambiguity, and represent a north star for everyone to aim for.
Always have a human fallback option, allowing the user to express “I’d rather wait and talk to a real human, make this robot thing go away”.
Start with defining key user intents that you believe your chatbot will encounter and the ones you should support. The scope is key here. Carefully define what you should cover and what you will not.
Every interaction with your bot does not have to be conversational. Some interactions are better with Graphical UI than with Conversational UI.
Researchers have proven that humans tend to anthropomorphise machines in a natural way. That’s why investigating the structure and process of social interaction between humans can enable better conversational interfaces.
Make sure you pay close attention to the language and vocabulary your chatbot uses to interact with users, as well as the info you can convey in association with the messages it sends — your chatbot’s “fist”, or meta-messaging. The UX design choices you make here will shape your user’s perception of its persona, which sets the tone for the entire interaction.
It may sound far-fetched, but remember that people are neurologically hard-wired for the rapid onset of boredom when repeatedly exposed to the same stimulus — so it’s this element of unpredictability in your chatbot’s linguistic repertoire that’ll keep your customers engaged.
Creating a consistent voice: We started with a high level voice guideline with personality. As mentioned, we tested different tones from more human to more bot-like. We found that sounding robotic increased efficiency. People had a lower expectation of what it can actually understand. When the bot sounded too smart and too much like a human, people pushed its limits by saying things that had nothing to do with a request.
Building trust: People will learn to trust the bot if it’s consistent, accurate, looking out for them, responsive, transparent, and smart. Biggest hurdle is a smart AI, but it’s not the only trust signal we need to design for. Sentence structure, images, context, and nagginess all are design problems.
Immediately provide some value or utility to the user. This drastically increases the likelihood that the user will continue to use the bot and not just forget about it. No one wants to answer 5 questions without knowing what they’re going to be getting.
You’re never done tweaking your bot, look at conversation logs and see how users are responding, come up with a small change and try it out.
Taking turns with short responses, like any good conversation, keeps customers engaged and holds their attention better. While a paragraph of text may be quick to skim and read, the nature of voice interaction is linear — you can’t skip around the way you can with a piece of text, and long, complex sentences can overload the user. If responses are too short though–such as a simple “yes”–the conversation becomes too mechanical. Finding a balance and testing different phrases that align with the persona is key here.
We all love stories. We’re born for them. Stories affirm who we are. We all want affirmations that our lives have meaning. And nothing does a greater affirmation than when we connect through stories. It can cross the barriers of time, past, present and future, and allow us to experience the similarities between ourselves and through others, real and imagined.
Iteration and usability testing
Getting confirmation is imperative in bot interactions. Designers should build interactions with the assumption that errors will happen early and often, given the ambiguity and impreciseness of most human dialogue. Ask for confirmation from the user for any critical step in an interaction.
For a start, companies that are new or have a smaller digital footprint can benefit from things like forums or even competitor reviews to get a better sense of the users in their industry vertical. And for more established companies, customer service logs and app reviews can be invaluable for learning what users think about specific products.
Here you can find a short directory of 15 tools and websites where you can test your chatbot.
Here’s a short list of bot developing tools, ranging from NLP solutions to bot builders, analytics and more, which are already available and can serve you. The list is divided according to categories (some tools may appear under more than 1 category) and includes a short explanation about each tool and a link for further information.
The purpose of this guide is to help you create a bot from scratch, from the first thought of its existence to its launch in production. This article focuses on the project management and bot building more than on the coding and framework implementation.
Measuring success
Traditional metrics like DAU / MAU and analytics tools like Google Analytics or Mixpanel work well for websites and mobile apps, but the unique conversational nature of chatbots requires a different perspective on performance.
“A long conversation doesn’t necessarily mean an engaged user,” says Ilker Koksal of Botanalytics.
Think beyond acquisition and towards activation. What is one action you think all your successful chatbot users need to take? Measure that KPI.
It is important for startups to properly instrument the data they track so that they can get a handle on the true health of their business. If they track only the vanity metrics, they can get a false sense of success. Just because a startup can produce a chart that is up and to the right does not mean it has a great business.
Chatbots for Marketing
Being an early adopter of a new channel can provide enormous benefits, but that comes with equally high risks. This is amplified within marketplaces like Amazon. Early adopters within Amazon’s marketplace were able to focus on building a solid base of reviews for their products — a primary ranking signal — which meant that they’d create huge barriers to entry for competitors (namely because they were always showing up in the search results before them).
Giving people a clear and precise expectation of what you will be sending, how often, and offering them the opportunity to choose whether they want to receive this type of content and how frequently. Give the users as much control as possible.
Marketers can send mass messages to specific target audiences and drive traffic to their bot. That is exactly what KIA did. It made sense for such a big company to advertise with bot messages before anyone else was doing it at that scale.
Further reading
Common sense, an old observation goes, is uncommon enough in humans. Programming it into computers is harder still. Fernando Pereira of Google points out why. Automated speech recognition and machine translation have something in common: there are huge stores of data (recordings and transcripts for speech recognition, parallel corpora for translation) that can be used to train machines. But there are no training data for common sense.
Proper conversation between humans and machines can be seen as a series of linked challenges: speech recognition, speech synthesis, syntactic analysis, semantic analysis, pragmatic understanding, dialogue, common sense and real-world knowledge. Because all the technologies have to work together, the chain as a whole is only as strong as its weakest link, and the first few of these are far better developed than the last few. The hardest part is linking them together. Scientists do not know how the human brain draws on so many different kinds of knowledge at the same time. Programming a machine to replicate that feat is very much a work in progress.
Abstraction is an essential property of software: the app you are using to view this piece is an abstraction layer above some operating system that knows how to read files, display images, etc. and this is an abstraction above lower level functions. Ultimately there is CPU-level code that moves bits — the ‘bare metal’.
We’ll try making a simple & minimal Neural Network which we will explain and train to identify something, there will be little to no history or math (tons of that stuff out there), instead I will try ( and possibly fail ) to explain it to both you and I mostly with doodles and code.
Be Credible. Ironically, it is the one thing that everyone thinks they have, but most do not! Most messages fail due to lack of credibility. The best way to address the issue, is via a ‘Try Before you Buy’ approach.
Turns out, most real life complex tasks where we are trying to optimize for an output that is a function of a large number of known/unknown, related/unrelated input variables, writing specific “if ___ then ___ unless____” logic is highly inefficient.
It seemed like the usefulness derived from the different tools was dependent on the existence of a purpose.
In our earlier articles we described the future, as we saw it, — an epoch of the Internet of People (IoP) 1.0 and 2.0 as well as the progress of the Convosphere. Now we’d like to tell about the current achievements connected with the development of the robot-assistant (chatbot).
People want advice and are willing to pay for good advice. As bots become more sophisticated, I expect people to be willing to pay to have conversations with the bots that can help them with various challenges in life.
Bots and AI for Good
The study offers a fascinating finding: machine learning — a future frontier for artificial intelligence — can predict with 80–90 percent accuracy whether someone will attempt suicide as far off as two years into the future. The algorithms become even more accurate as a person’s suicide attempt gets closer. For example, the accuracy climbs to 92 percent one week before a suicide attempt when artificial intelligence focuses on general hospital patients.
X2AI describes its bots as therapeutic assistants, which means that they offer help and support rather than treatment. The distinction matters both legally and ethically. “If you make a claim that you’re treating people, then you’re practicing medicine,” Rauws told me recently. “There’s a lot more evidence required before you can make that claim confidently.”
We have the opportunity to create an interface that is able to help disabled people (blind, deaf, everyone) in every medium (computer, phones, etc.).We as designers, developers, business people, and everybody involved in the Chatbot ecosystem should take this on consideration. We should not only focus on creating something just new, but on creating something that is new and accessible for everyone.
But it’s not just tech giants like Facebook, Instagram, and China’s up-and-coming video platform Live.me who are devoting R&D to flagging self-harm. Doctors at research hospitals and even the US Department of Veterans Affairs are piloting new, AI-driven suicide-prevention platforms that capture more data than ever before. The goal: build predictive models to tailor interventions earlier.
Errors, it seems, do not entirely deserve their bad reputation. “There are many, many natural processes where noise is paradoxically beneficial,” Christakis says. “The best example is mutation. If you had a species in which every individual was perfectly adapted to its environment, then when the environment changed, it would die.” Instead, random mutations can help a species sidestep extinction.
Good examples of bots
- Poncho — all about weather. An example of a chatbot with a great personality.
- And Chill — gives you movie recommendations. Quick, efficient and useful.
- Dankland — makes memes out of pictures. A genuinely fun bot to play around with, simple and effective.
- Bearhug — lets women track their periods. A great use case of a chatbot that is superior over apps and other solutions.
- SkyScanner — search for flights on messenger. Extension of the service to messenger, does the job done, especially a convenient use of a chatbot for alerts.
- Instant Translator — Quick translation to more than 18 languages on the go. Very simple to use.
- BFF Trump — Gives you all the crazy things that Trump has said. Entertaining, simple and was a perfect use case for the election period.
- Ebay ShopBot — beta chatbot of Ebay to make the search process easier and quicker. Has a lot of artificial intelligence integrated and is getting better over time.
- Chatbot that overturned 160,000 parking fines now helping refugees claim asylum
The original DoNotPay, created by Stanford student Joshua Browder, describes itself as “the world’s first robot lawyer”, giving free legal aid to users through a simple-to-use chat interface. The chatbot, using Facebook Messenger, can now help refugees fill in an immigration application in the US and Canada. For those in the UK, it helps them apply for asylum support.
When Visabot went live last November, the Facebook Messenger-based artificial intelligence attempted to simplify the US visa application process and help many people skip the fees associated with a visit to an immigration lawyer.
The House Museums of Milan is a group of 4 amazing historical homes in Milan. When they launched a new project that aims to motivate people to visit the four homes through a single itinerary they decided to introduce some gamification into the process so as to attract younger audiences.
Want to know more? Tweet me @tomhewitson,send me an email at www.tomhewitson.com or subscribe to my newsletter.
Also, thanks to Rokas J for helping me put this together and the push to get it published.
