Designing Conversational Experiences

This high-level view of how to design conversational experiences is the second in a series of linked posts exploring the future of Experience Design and a follow on from the previous post on The Future of Conversational Design. These posts are a work in progress, the next five years is going to see a series of huge changes, many of which are just coming into view and there’s no right answer, so please get involved, add comments, refine my thought, disagree, etc.

High Level Considerations

The experience, UX and service design communities have been working with conversational interfaces for some time now. However, a lot of new learning and best practice has emerged over the past 18 months as conversational experiences have rolled out to more mainstream audiences. This post brings together some of the most interesting cases studies and best practice guidelines, as well as links to useful conversational design resources.

Four high level best practice considerations

There are a range of high-level considerations for experience designers working on conversational experiences, however, there are four that stand out as starting points for those who are new to conversational design:

  • clearly define the use case
  • set user expectations and build in appropriate affordances
  • design personalities
  • think ‘beyond’ text and voice

Clearly define the use case

Though 2016 was the year of the chatbot, many of those developed where either small POCs or campaign lead PR stunts, most of which aren’t great examples of conversational design, or haven’t been tested with large numbers of users. As we’ve seen reliability remains a big issue for voice and chatbot users, with fail rates of up to 70% on Facebook messenger. This means that the first task for experience designers and strategists is to consider the use case carefully before deciding on a conversational design solution. So for example, placing a chatbot in a first-line customer service context with a 40% fail rate will almost certainly frustrate users and add costs. However, reliability may well be less of an issue if bots are used as part of a marketing campaign or platform where they’re not expected to perform important tasks for frustrated users.

I’ve picked out five areas where conversational interfaces are already adding real value. There are more, and these specific examples are for the most part early implementations, so aren’t perfect, however, the following applications show that there are already a wide range of implementations where conversational experiences can add a great deal of value.

Internal colleague experiences — e.g. training, onboarding or information sharing, for example Growthbot on Slack which allows team members to use chat to figure out what’s happening with internal analytics, e.g. Google Analytics, Hubspot and a dozen other systems and data providers to provide quick and convenient answers to common questions, e.g. how many uniques did the site have today?

While any error or failure is obviously going to impact the user experience, users of internal systems tend to have higher tolerance levels, and there is no external reputational damage.

Learning — Duolingo’s chat bots let users interact in French, German and Spanish with a chatbot, helping them learn the language in a more ‘natural’ and conversational way. In addition, what makes Duolingo’s approach stand out is that the company has created more than one bot, each of which has a different personality — Chef Robert, Renée the Driver and Officer Ada all give subtly different answers and take conversations in different directions. This allows users to test different skills, but also engage on a much more emotional level, which in turn helps with.

Again, fail rates don’t have such a large impact on the user experience (particularly as users may not notice them!)

Recommendation and suggestion — chat and voice applications are great for recommendation and suggestion, in part because there is no ‘wrong’ answer. Obviously the quality of the recommendation engine is important and better recommendation will drive increased usage, however, there errors bring with much less cost than they do in customer service situations. The H&M chatbot is a good example of a well designed and considered recommendation application, acting as a personal shopper for users.

Recruitment — Mya is a recruitment bot is that lets recruiters automatically prescreen candidates, so they can focus on interviews and closing offers. Mya asks questions; responds to questions; delivers progress updates and gives tips and guidance to candidates, and can talk to thousands of candidates simultaneously through SMS, Facebook, Skype, or chat. It can also help rank and sort candidates for human screening. There are obviously potential issues with this approach, and it’s unlikely anyone will use Mya to hire their next CEO right now, however, for companies hiring a lot of people or dealing with a lot of applications, it can, as their reported results suggest, add real value.

Customer Service — having said that high error rates currently make customer service a difficult use case for chat and voice, there are plenty of large organisations successfully implementing them. For example, in 2015, KLM was one of the first businesses to develop for the new more open Facebook Messenger platform, allowing users to confirm booking, get boarding passes and use it as a customer service channel. KLM passengers can now use it as a recommendation service, asking it for directions to the nearest shop, restaurant, shop ATM or transport hub. Soon after launch, KLM said 1.7 million messages have been sent on Messenger by over 500,000 people.

In the next section we’ll look at how managing expectations, building in affordances and fail safes can help mitigate issues around error handling for front line customer conversational experiences.

Expectations, Affordances and Failsafes

As well as carefully considering the use cases for conversational interfaces, experience designers also have to build in appropriate affordances, so that users understand the limits and uses of the systems. This makes for a better experience as users appreciate what the system can do for them, rather than complain about what it can’t.

Intercom, a pioneering product company in messaging and AI, has some very clear advice learned at the sharp end of implementing chatbots and other conversational experiences:

  1. Use it sparingly — it’s currently reliably good for a small range of relatively simple task.
  2. Keep it very simple — don’t try to make interactions too rich, keep conversations short, and simple with clear scripts and flows.
  3. Heavily structure conversations — offer few options, avoid open-ended questions
  4. Always have a human fall back — no consumer facing AI can currently deal with complex customer service issues. It is best used to complement and triage queries.

Designing Scripts and Personalities

New more human and natural interfaces require new skills. Chatbots and virtual assistants will increasingly require distinct personalities that will have to be designed. Experience designers now have to consider a new range of questions, e.g. how casual or informal is the tone of voice, does it line up well with the copy of other touchpoints, and what about the gender of the bot,w hat gender’s going to be best to engage your specific audience, or is it gender neutral. How closely does the personality of any bot reflect the personality of the brand? If you have more than one bot, then do they have different personalities — e.g. a more serious one for customer service and a more lighthearted one for brand marketing.

Think beyond Text and Voice

On the face of it this advice may seem a little Zen, however, chat interfaces have evolved a long way beyond simple text, with emoji, gifs, videos and images all increasingly used in interactions on chat platforms. And, as we’ve seen, chat platforms are now access points to a much wider set of content and applications than just chat.

As a consequence experience designers need to consider both inputs and outputs to chat, and outputs to voice.

A perfect example is the recently released Dominos chatbot on Facebook Messenger where the posting a simple pizza emoji is enough to initiate an order for pizza.

Experience designers need to consider how different types of input and output can work through any experience, e.g. does a chatbot send back pictures of solutions to common problems faced by people assembling flat packed furniture, rather than attempting to help in text. Can you use gifs instead of videos?

Designing Conversational Experiences — Key Links and Reference Material

There’s already a significant amount of practical advice for experience designing conversational experiences:

Chatbots Magazine is a good starting place, and in particular their Beginners Guide to Chatbots.

There’s also lots of great tips on conversational design in this Google guide from their conversational design team.

There’s also always lots for experience designers to learn from the developer platforms, which often have introductions to functionality and interactions — some of the bigger ones include:

Facebook Messenger Developer Platform

We Chat Developer Platform

Google Actions is the start point for designing conversational interactions and applications for Google Assistant

Alexa Skills is a similar starting point for Amazon Alexa

Understanding and measuring user interaction with bots

An in depth guide to the logic and technology used to build a chatbot

15 of the best Slackbots

Tom Hewitson’s blog dedicated to Conversational Design

If you’ve got a link to any other related resources, it would be great if you could add them in the comments below.

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.