Earlier this month I was lucky enough to attend Interaction17, with a couple of my RMA UX design colleagues, Mei and Dario. This year more than 1,000 designers travelled to New York City for the IxDA’s 10th annual gathering.
Having attended two of these IxDA events over the years, I always find it interesting to see the changes in what the community is getting excited about. This year the trend that stood out to me – and featured in at least five presentations, rather than one last year – was Conversational UI. When I returned to London I shared my key learnings with the RMA design team and the delegates at the IxDA London meetup Interaction17 Redux.
Conversational UI (CUI)
The recent improvements in voice processing and advances in AI, which have made computers and smartphones better at understanding what users really mean, has resulted in the dramatic growth of Conversational UI (CUI).
A good CUI takes some of the good features of talking with a real human. Rather than clicking on icons you interact with your device, by just telling it what to do. It is often used to help automate the kinds of tasks you would usually do on your own, like making a dinner reservation, adding an appointment to your calendar or obtaining information.
There are two different types of Conversational UI:
- Voice driven interfaces (ie Siri, Echo, Cortana) where the input and often the output is primarily through old, but new, medium of ‘voice’.
- Messaging platform based interfaces, with chatbots, such as M in Facebook’s Messenger and Slack’s Slackbot. These are an increasingly significant venue for digital engagement. By 2018, apparently 2 billion people will be using messaging platforms.
In both cases of CUI there are two types of Bots –
- Transactional Bots
- Conversational Bots
Transactional bots are like telephone voice experiences of old. The conversations are restricted by a ‘script’, but the focus is not on the conversation itself – but the efficiency and speed with which it allows you to achieve your goal.
What’s the design process?
The rules of cooperative conversations described by the philosopher and linguist Paul Grice also apply to transactional bots:
▪ Quality – say things that are true
▪ Quantity – say only as much as is needed
▪ Relation – keep things relevant
▪ Manner – get to the point, and avoid obscurity
Greg Vassallo, the UX Design Principal at Fidelity, recommended the following practical steps to designing for transactional bots:
- Role play workshops – Play out the interactions with one person representing the bot and the other a human, then take transcripts and draw out the dialogue structure.
- Map out the dialogue flow – Brainstorm keywords, list out user intent, identify the possible dead-ends and ‘help’ opportunities.
- Brand experience – Consider how your bot’s personality fits with your brand and how quickly your bot responds to questions. If the response is too fast it will appear spooky and put the user under pressure.
Facebook’s Content Strategist and Storyteller Elena Ontiveros went into further detail during her presentation. She emphasized that copy writing is more important to the user experience of Conversational UI than the visual design.
Elena advised delegates to think carefully before attempting a ‘personal’ style with Transactional Bots and that obviously ‘being a bot’ is less confusing and likely to annoy the user. Having said that, she also stressed the importance of empathetic and emotional content; writing a minimum of four variations for each message to keep the content fresh, and to read the content out loud… to make sure the cadence and tone ‘feels’ right.
Greg Vassallo gave a useful presentation on the different tools available for designing for CUI. The tools included Bot Society and MotionAI. Bot Society enables you to design, preview and export your chatbot as you go. But according to Greg, MotionAI “visually builds, trains and deploys bots to do just about anything.” The graph based approach helps you to build a conversational UI dialogue structure and then report on how people are using it.
- Elizabeth Allen, UX Researcher shared her two examples of Shopify’s Transactional Bots. Kit is a service for Shopify merchants that makes suggestions about typical marketing tasks it can do for them.
2) Messenger is a service for Shopify merchants to use with their customers, to let them shop where they are the most – in chat.
3) Slackbot is Slack’s own internal help bot. As well as helping Slack users, it can be customised to automatically respond to a teams frequently asked questions – eg the office wi-fi password. It also has some alternative fun uses – last Christmas we used Slackbot with its Secret Santa plugin, to randomly and anonymously assign and notify people who to buy gifts for.
Conversational bots differ from transactional bots as they interpret what you’re saying and attempt to ‘intelligently’ respond with the relevant dialogue to achieve the ‘bots’ goals, be it getting something specific done for you, or just having an entertaining chat.
What’s the design process?
Another presentation by Whitney French from WillowTree focused on how to create a conversational bot with (some) emotional intelligence. She recommends that you:
- Shape a bot mission statement – A concise, concrete statement of your bots purpose and value
- Get the right training data - You have to make sure that you have the right data since you’re not designing behaviour, you’re training an AI
- Cover the edge cases – make sure you cover all the many things that can go wrong. Leverage all the data sources you can get your hands on (e.g. call logs)
- Build a bot workshop –
- Set up several pairs to role play ‘bot’ and human players
- Run chat via a chat tool such as Slack
- Create cards as you go, create cards with the different entities and parameters (e.g. for movies: titles, times, places)
- Look to find humour to get out of tough spots in conversations
The tools for conversational bots are development tools. Each of the big players has a platform you can leverage to build this kind of experience:
Facebook’s Wit.ai is one such AI tool kit, used by Facebook, which allows you to rapidly build open-ended conversational UIs. The AI maps the possible outputs to inputs rather than giving specific outputs, which enables the conversational bots to be far more flexible and dynamic.
Google’s api.ai tool builds conversational experiences for Google Assistant, by helping the user complete three steps – 1) Design interaction scenarios and conversation flows; 2) Connect your business logic and fulfill user requests; 3) Launch for Google Assistant users on Google Home.
IBM Watson Conversation can receive the users input from any user interface (ie mobile app, voice interface or robot), it then interprets the intent and gets the information it needs. Depending on these factors IBM’s Conversation can then respond in a limitless number of ways – from answering a question, purchasing tickets, updating account information, or placing orders.
Microsoft Bot framework is made up of three components – the Bot Connector, Bot Builder SDK and the Bot Directory to give users all they need to build and connect bots. As the name suggests the Bot Connector connects your bot to multiple user interfaces, such as text, email (Office 365), Skype, Slack etc. The Bot Builder SDK is an open source SDK hosted on GitHub, which gives you everything you need to build conversations. The Bot Directory lists all the bots you can connect to.
Microsoft launched Zo on Kik, a messaging app in December 2016. Unlike transactional bots this app has abundant personality. Zo can answer questions and respond to prompts, while using teenage slang, and emoji – ‘she’ is targeted at millennials. But with so many possible outputs Zo seemed to get easily confused and go off tangent. I personally didn’t really gel with Zo; she just didn’t seem to understand anything I said… but then perhaps that says more about me then Zo (and yes… I’m not a Millennial).
What does CUI mean for business?
The technology is now available for CUI to take off. It offers businesses the potential to reach more people than ever as it resolves the app discoverability problem and is easy to use, which makes it more cross generational than any other communication channel.
However, in reality how and why would businesses invest in conversational UI (other than keeping up with a new trend)? Here are 10 ways that businesses can either make money, save time or improve customer experience with CUI:
- Customer service: Help and support desk chat is currently the most obvious and popular business case for CUI eg @slackbot
- Data processing: Industries like the legal sector can use chatbots to search, read, interpret and summarise large amounts of data 10 million times faster than people eg Ravn ACE
- Finance advisers: Chatbots can help users with simple but common issues, such as lost debit cards and locked PINs eg RBS’s Luvo
- Accounting: Users can track expenses and manage finances by just having a conversation via messaging platforms like Facebook’s Messenger and Slack. Eg Sage’s Pegg
- Medical adviser: Similar to the UK’s ‘111’ service, users can access medical knowledge, get advice, and be directed to the nearest medical facility, saving patients time and healthcare services, like the NHS, money
- Advertising: As chatbots replace some of the functions that people normally use a search engine for, advertisers will pay for the privilege to access users
- Subscriptions: If bots can save users the time that they’d normally spend on mundane tasks, they will be willing to pay a subscription cost
- Commission: Retailers could pay the bot-maker commission for helping to sell products
- Personal concierge: The next evolution of Siri and Cortana will be bots that will be able to organise your emails, diary, appointments and access to other bots.
- Online orders: Rather than completing online ordering forms, users can chat or type to a bot to make their order eg Tacobot.
What does CUI mean for us designers?
CUI is still very much in its early stages, which means that at the moment there is still a lot of potential for businesses to get it wrong. With 93% of communication being non-verbal (55% made up of body language and 38% tone of voice) there is a lot of room for misunderstandings through CUI. Microsoft’s controversial Tay is one example of how badly it can backfire. Bad conversational UI is damaging. If it can’t be done well – don’t do it at all.
The key to good conversational UI is good copy writing and the building blocks and tools are developmental… so is conversational UI the end of design? Obviously I don’t think so. Here at RMA, we’ll be defining what role user experience and visual design will have in conversational UI.