Ambit’s Design Lead, Amelia Diggle, shares her top 5 tips for chatbot design.

After helping over 100 NZ organisations design chatbots — we’ve learnt a thing or two!

Here’s our 5 top tips for creating a great human-centred chatbot.

Top tip 1: Create a T-Shaped Chatbot

Just as we try to hire “T-shaped” employees — we’ve found it’s best to create T-shaped chatbots. The T-shaped metaphor is used to describe the skills and abilities of a person within a team. The vertical bar representing depth of knowledge and experience within a specific topic, and the horizontal bar representing the ability to collaborate and work with people in other areas by having a light understanding of their topics/disciplines.

Why do this?

Because you have no control over what your customers or employees will chat about it’s best to start off by lightly covering #allthethings — before giving the details on anything specific. And you don’t want to chat to a bot that can only talk about 1 thing, or a bot that only lightly covers everything #surfacelevel 💁‍

Some of our clients are starting to expand their T-shaped chatbots to become S, or N, or M-shaped! Meaning they are now able to have in-depth conversations about multiple topics and also interlink between these topics!

First impressions are so important. You want your customers first chat to be great — or they might not chat again. So by understanding entire customer experience first you can find out about all the topics you need to lightly touch, and then decide on which topic(s) you want to go deep with — and flesh out a detailed, helpful and meaningful conversation for.

Top tip 2: State that is is a bot, but talk like a human.

This one has been quite controversial. When a chatbot says that it’s a chatbot — it’s almost going against the turing test purpose of trying to trick a human into thinking a machine is a human. But the people of New Zealand have spoken, and we’ve spoken to a lot of you about this — people of all ages and backgrounds, and we all prefer honesty, and transparency — we want to know what we’re talking to. So even though our chatbots can chat like a human using the latest Artificial Intelligence — we still want to know what it is.

For example — Jules (ACC) and Coops (Co-operative Bank) both state that they are Bots — but quickly move on into natural conversation.

Top tip 3: Take turns leading the conversation.

This tip is about balancing how you use AI in your chatbot. Looking at which type of conversations work with ‘more AI’ vs scripting. Scripting enables the business/chatbot to lead the conversation, but NLP (Natural Language Processing) enables the chatter/customer to lead the conversation.

At Ambit, we simply conversations into 3 different types derived from an end to end customer experience — Acquisition and Retention, encompassed by Help/support at any point in the experience. We got to these simplified conversation types by looking at trends across all our clients conversational ecosystems — finding there’s common customer/employee needs in every organisation that fit into these 3 common buckets.

It won’t come as much of a surprise as to who should lead the acquisition conversations — the customer. One of Dale Carnegie’s top tips in his famous book “how to win friends and influence people” is to never talk about what you want — always find out about what others want — then talk about how what you have to offer could give them just that.

But once you have customers — when they are trying to get tasks done with the things they already have and use from your organisation — this is when you (the chatbot/business) can lead the conversation. You can lead your customer to quickly complete tasks and make the most out of your products and services.

Help and Support is more of a dance. Start by listening to the customer (using NLP/AI) to identify the issue, then you can lead the chatter to a resolution (using scripting). And sometimes it will be a lot more back and forth — as you troubleshoot to find out the exact issue and best resolution.

Top tip 4: Pick carefully where you ask for chatter feedback.

Humans don’t ask halfway through a convo, “hey, how’s this chat going?”. (although maybe we should then we might be better listeners….)

This tip is about qualitative feedback. Hearing exactly what your customers think and want is like being given a get out of jail free card for bad customer experiences. But when do you ask for feedback? At the end of every topic? After each question has been answered? Do you just wait for people to give it? Or do you ask every second person??

Where is it most useful to gather feedback?

We don’t have a one-size-fits-all answer for this, but we do know it’s important to pick your time carefully. Don’t interrupt the conversation flow and risk losing your chatter by asking for feedback, look carefully for useful places to ask and monitor your analytics to see if this affects your drop-offs.

An idea for new chatbots, when your chatbot is new and still learning more about what your customers want to talk about it will have questions and tasks that it simply cannot do — so instead of just saying “I’m sorry I don’t understand” — this is a great place to ask for feedback.

Top tip 5: Make it fun!

The joy of this user experience and interface is that it’s informal — so use it — it’s a conversation! Have as much fun as you can (within your brand guidelines and values of course 😎).

With crime.

And now you think I’m crazy. Keep reading a little bit more. After all — bad design is a crime ;)

Intelligence analysts have been analysing crimes for centuries — or at least since Sherlock Holmes. Digital is still new and we’re still refining how we analyse the different interactions and aspects of our digital experiences. So, I’ve taken a leaf out of the intel book and applied it to conversation design.

But why?: And who are you? You’re a designer, not an intelligence analyst?

I’ve designed user interfaces for intelligence analytics and detectives — and through this I learnt all about intelligence and crime analysis techniques.

This article gives you another technique to help you with the next phase of design thinking; Define. To quickly get clear on what each conversation topic is about (define).

Next article will be on Ideate — an activity to start ideating on how the conversation could play out.

In Part 1 of Winning Conversations we shared the Conversation Ecosystem. A visual tool to represent all the possible conversations you would like your chatbot to be able to discuss.

Define your conversation

The define phase of design thinking is all about defining the exact problem you want to solve. Before ideating on possible solutions.

Conversation problem analysis triangle (CPAT)

This is where crime comes in handy.

“For a crime or incident to occur, an offender and a suitable target must come together in a specific location without an effective deterrent. The figure below shows how these three elements: victim, offender and location, form the problem analysis triangle (PAT). When developing a description and understanding of a particular crime or disorder problem or individual, analysis should consider each of the elements of the triangle. Further, when developing solutions to that problem, the three elements should again be considered as a focus for activity.” UK Police college.

Makes sense, right?

Translating that to conversation design you get:

• Victim becomes the customer (we call them “chatters” in conversation design.)
• Offender become the offering. What are you selling, or what are they needing help with. Could be services or products.
• Location becomes context. When and where are they most likely to chat.

But that would’ve only catered to your external users — your customers. Digital employees can be designed to talk to both customers and employees. Let’s think about each conversation from both sides of the coin — create a Chatbot that can really talk to everyone. So, I’ve flipped the triangle and given you two victims — the customer and the employee.

Your conversation design tool-kit

Conversations are an art and take tact. Ambit arms you with the tactics, tools, and human-centred-design techniques to achieve a winning conversation design. So far we’ve shared:

• How to Understand the potential entirety of your Chatbot: with the Conversation ecosystem diagram. A visual representation everything your chatbot needs to be able to talk about.
• Then how to Define each conversation: with the Conversation Problem Assessment Triangle (CPAT). A tool taken from Crime Analysis (the PAT problem assessment triangle) giving you a way of developing a description of a particular conversation in preparation for ideation, by considering each of the elements of the triangle you define the conversation purpose.

Keep an eye out for Ambit’s next instalment of winning conversations on how to Ideate on your first few conversations to cater to these different contexts, people and topics on a small chat screen.

Conversational user experiences (UX) take a lot of work. There’s a reason many state “I’m like a child and just learning”, whilst endeavouring to deliver adult grade conversation within months or even weeks of birth. Even for a basic chat to work well requires balancing NLP with scripting, simplifying steps to complete tasks, capturing data in appropriate places and balancing the ‘creep’ factor of AI and big data. The field is rapidly evolving and the rules along with it, this article intends to give you some powerful guiding principles to get underway.

A lot can go wrong (there’s even a website just on chatbot fails!). Designers must take careful consideration of each conversation — and we want to pick the most beneficial place to start to build, test, and learn from.

But #1: which bit to build first?

Introducing — The Conversational Ecosystem!

Conversations are complicated. They ramble and weave, change context randomly and cover many different aspects and topics. Replicating this in a machine is not an easy task — one of the many reasons why AI has taken 70 years from the “Turing test” to reach this point.

Conversational Ecosystem diagrams give you a high-level view of #allthethings your customers and employees need to chat about with your organisation. We created the conversational ecosystem to quickly get to a holistic view of a businesses conversational opportunities and how they relate to each other.

Visualising conversation relationships enables you to make informed decisions on where you want to start designing your chatbot. They can also help you design the first conversation better, by understanding what it will need to link to or what other conversations people might need to have next. For example, let’s say a travel Chatbot has the topic of booking plane tickets. The topics within that topic would simply be the methods of booking a ticket. For example, getting a recommendation from the bot or searching the bot for a specific destination. Once you have this mapped out, it is far easier to see potential linkages and value adds.

But #2: conversation design shouldn’t be rocket science!

Ecosystem models are used in science to understand ecologies, so it’s not rocket science! Designers have been using them across multiple industries to understand systems and scenarios. Architecture is often represented in this way too. Just as you need to design a house so that people can move between the rooms in a practical way — you can use ecosystem diagrams to help you design your conversations and content in this way too.

But #3: why not just design the conversation steps straight away!?

Your Chatbot/Conversational UX is another window into your business — and people are going to ask it #allthethings. So without understanding how all the conversations fit together, you can’t set yourself up for success from the beginning. “Understand” is the first step in the ever-popular and practical process of design thinking.

“By using design thinking, you make decisions based on what future customers really want instead of relying only on historical data or making risky bets based on instinct instead of evidence.” Ideo.

Understanding how converations relate and link to each other with ecosystem thinking sets you up for long-term success. These diagrams empower you to choose; design the light touch (MVP) for each topic, focus on one specific area, or craft a specific path through. This enables you to more confidently design the first few conversations to grow into a fuller complete conversational UX.

An alternative approach would be to do a customer journey map — design conversation steps or touch-points — whilst a very useful aspect of the design process, this linear view doesn’t give you that birds eye view confidence before diving into define and design the conversation details.

But #4: I don’t know what the key conversations are?!

Here’s some ways to identify your key conversation topics:

• Review existing web-chats or phone call data for the top FAQs
• Look at your top web search terms or google search terms
• Brainstorm the key tasks customers and employees need/want to be able to do with your organisation

Once you’ve identified the key conversation topics/tasks, you can use Information Architecture design techniques to identify the key topic groups and their relationships — such as card sorting.

But #5: I have very different types of customers!?

Good news — conversation UX designs can be adapted for the different personas/user profiles/customers/segments. It’s easy to change the tone or tweak the content! Ambit’s platform enables you to easily personalise conversations for your different types of customers/employees. For example purchasing conversations can be used for consumers and businesses — just change the product info, links and tweak the tone for your different audience.

The power of an image.

Visually representing conversations gives you a powerful way to get started designing and building your conversational user experience. Like any project — start with the end in mind. Understanding your customers, what they want, and how they communicate with you is key to winning conversations.

If you think of any “if’s and but’s!?” please ask away — or get in touch.

And keep an eye out for Ambit’s next instalment of winning conversations on how to define your first few conversations.

This story is published in The Startup, Medium’s largest entrepreneurship publication followed by 295,232+ people.

Subscribe to receive our top stories here.

Winning conversations: how to design conversations that influence people was originally published in The Startup on Medium, where people are continuing the conversation by highlighting and responding to this story.

Trust. The key to a perfect relationship. In a world where knowledge is power, consumers are more and more concerned with their data. (Emphasis on the ‘their’).

The data futures partnership issued a report in August this year with guidelines for trusted data use, highlighting the questions New Zealanders want answered when it comes to the security and use of our personal data on the Internet.

Organisations need to listen to what the community has to say about its proposed data use and to make changes if the community is not comfortable.
The Data Futures Partnership

So what are businesses really doing with our data? Selling it? How much of it do they have? How do I know what they have? Where is it? Can my crazy ex-husband access it (fingers crossed I never have one of these!)? Is it safe? How much is it worth? Why do they ask me about my [insert unnecessary form question]? Is it anonymous? Can I delete it? What if it’s incorrect?! What if my data is being held captive and being excluded from the important strategic decisions it should be influencing!? What is data? Who am I?

So.
Many.
Questions.

And, you know who’s good at answering questions? A chatbot.

You don’t have to trawl a complicated Google search flipping through articles or ask a data issues lawyer. You just go on to your messaging app (that you’re already on) and ask a bot!

For businesses using chatbots, chatbots create an opportunity for both the business and the customer. Customers can ask #allthequestions they can possibly think of to a Chatbot, and at the same time, the business can listen and learn from the actual voice of their customer. No assumptions or complex analysis need to be taken.

From the horse’s mouth

The real “so what?” is when your business can contextually answer these questions and remove friction from the customer experience. Human behaviour changes online, (read The Cyber Effect by Dr Mary Aitken if you don’t believe me!). Chatbots create a forum for customers to ask the questions they wouldn’t do in any other environment — and your business will be able to learn about the context in which they are asking these questions. For example, Ambit’s Chatbot Platform allows you to iterate, improve, evolve, adapt, pivot, and innovate your conversational customer experience from incoming messages.

Use cases abound for businesses that can solve a huge problem like the privacy concern that New Zealanders have above. Think FAQs — businesses are using chatbots for them so they no longer have to rely on call centres answering the same 80% of queries every day.

Perhaps it won’t be from the horse’s mouth, but from the Chatbot’s mouth one day.

Ambit’s Chatbot platform empowers your business to not only answer these important questions about trust and data, but enables your customers to choose what they want to share. And design customer-centred conversational experiences for capturing customer data with transparency. Check out Ambit here for more weekly content on Chatbots and conversational AI.