Learn about the conversation design process, including understanding requirements, tech considerations, identifying user personas, intent recognition and more.
We covered how conversation design can make or break chatbots in an earlier post. Here, we go into the process involved in the conversation design process. including gathering the requirements at the start, creating the basic conversation design flow and adding more detail to cover all possible scenarios and exceptions.
The first step in designing conversations involves understanding the requirements of the bot. The technology available, operational considerations and stakeholders, the user personas to address and the bot’s own persona are key factors that affect conversation design.
Before commencing a chatbot implementation project, it is vital to take stock of what technology resources are available. This includes the technology team and their specialties, platforms available, integrations possible and channels where the bot will be active among others.
Generally the tech team involved in a chatbot project comprises front-end and back-end developers, data scientists and even an AI expert. This also depends on the industry, type and size of the organization as well. For example, large technology companies may find it easier to create small teams within them that focus on innovation projects like chatbot. In contrast, a healthcare firm or an insurance company may not have a core tech team that can spend their time on innovation projects and may have to outsource their chatbot development to a specialised vendor.
Chatbots can be deployed using a variety of platforms and specific installations will have to consider which ones are available. Microsoft Bot Framework, Google Dialog Flow Bot landscape. Microsoft Bot Framework, Google Dialog Flow, IBM Watson are just a few of the many solutions available in a growing landscape. It is important to include the developers and engineers in the planning process to figure out which ones are most relevant, what they can and can't do for your project and how they will affect the conversation dialog design process.
A chatbot on its own might be able to automate conversations to a certain extent but its value will be limited without integrations to other platforms based on the use case. For example, a healthcare institution will find value in connecting the bot to their electronic medical records to be able to assist patients with queries about their medical history and hospital staff with details about their treatment history. In an insurance firm on the other hand, integrating the bot with policy management software and CRM systems can help customers and policy holders with more personalized service and recommendations.
Chatbots can be deployed across numerous channels today. They could be sitting on a company’s website helping visitors with FAQs and purchases or on Facebook, automating responses to common queries about opening hours, location and directions. They could also be in the form of a WhatsApp Business API enabled chatbot to help businesses scale real time communication with prospects and clients or on other mobile apps. Each of these make a big impact in the way conversations are designed.
It is also useful to decide at the outset if and how a user of the chatbot needs to be identified and when and how they should be handed over to a live agent. In some cases like a general FAQ chatbot, anonymous interactions are reasonable but for more transactional, sales conversations, it makes sense to authenticate and identify them before proceeding with personalized responses. For these transactions and sales interactions or when the questions get more complicated in nature, there needs to be a sensible handover to a live agent. Keeping these scenarios in mind can help craft the relevant dialogues during the conversation design process.
Like any technology project, the design of a chatbot has to serve a bigger purpose within an organization. This requires thinking about how the bot will help a business achieve its objectives.
For example, a bot may be created to help a SaaS product help website visitors or customers with trial sign ups and onboarding. Or it could be a travel insurance firm helping website visitors compare products, understand policy terms and check out in a self-service experience.
Even within an organization, there could be individual business units and departments who will stand to gain the most out of implementing the bot. Customer service teams, sales and marketing, IT support and HR are all departments whose functions can be improved with the implementation of a chatbot.
This also means that various stakeholders will also be involved in figuring out the direction of the bot's development. Some of the most common roles in a chatbot development team include Product Owner, Head of Customer Experience, Head of Digital, Innovation Manager or even the CEO himself in some cases. Knowledge Management and Internal Communications personnel are also critical in bot projects, as they are often in charge of managing and maintaining internal wikis, communication logs and reference content that the bot will use to train.
Lastly, the specific KPIs set out also play a big role in dictating how the conversation design process will progress. Some of the common metrics that are monitored in chatbot projects include
In the last case, the idea is that when the total interactions with the contact centre staff goes down, the duration and quality of each interaction goes up. This signals a more focused service where high value conversations are common.
In designing any conversations, you need a good understanding of whom you are writing for. This holds true for creating website content, sales conversations, voice bots and chatbots. To help arrive at this understanding, it is important to create a user persona.
A user persona is a semi-fictional representation of a specific segment of your target audience. It is semi-fictional because it is informed by real data and not by blind guesswork or stereotyping, but at the same time not a 100 percent accurate mapping of every individual who may fall in the audience segment.
User personas help conversation designers empathize with the audience and in creating dialogue that will most resonate with the audience. Often the data used to create personas may already be available in your organization, in the form of persona research undertaken by your UX or marketing teams. Look for these key elements when arriving at the user persona
For example, an example user persona for an insurance firm selling life insurance products could take the following form:
Alex, a 39 year old IT Engineer, living with his wife Sandra, 38, with two daughters Ivy, 7 and Tracy, 4, in Singapore. Alex has a yearly income of SGD 80,000 while Sandra makes SGD 96000 a year and both are graduates from a recognized university.
Alex wants to ensure a financially secure, healthy and happy future for his family. He intends to work until he is 60 years old, when he wants to have saved up enough for living a simple but comfortable retired life with his wife and hopes his children will have grown into stable careers by then.
Alex is familiar with chatbots and has interacted with a few while browsing websites, ordering products from online stores and delivery apps. He found it useful for basic queries but not for more complicated transactions.
A bot persona is the character behind the chatbot. It is what gives the bot its personality and differentiates it from a monotonous conversation with a robot. Bots with a clear persona can help your brand establish trust, consistency and likability with your audience and users. Deciding on a bot persona involves thinking about the following aspects of the bot.
Next, you need to decide the use case for your bot – what the bot will help you and your users accomplish. This could be different for different businesses. For example, a healthcare institution will a chatbot helpful in triaging patients and prioritizing patient screening or a travel and tourism firm could use a bot for providing weather updates.
Even within an organization, a chatbot could have multiple use cases. For example, an insurance firm might use a bot to bot provide self-servicing for customers and policy holders and for agents to look up key information on product updates, company announcements, promotions and definitions and terminology.
Within each use cases, the bot will be asked to perform different tasks. For example, a bot that helps order food will have individual tasks like ordering breakfast, lunch, dinner, booking restaurants, confirming orders, ordering cabs and so on.
From the bot’s perspective, the use case is often called an intent. Thus the first thing that the bot needs to do in a conversation between a human and bot is to recognize this intent. Only if the bot understands what the user wants to accomplish can it deliver the right dialogues to get the tasks done.
Once the user and bot personas and the use cases are decided, it is time to start creating sample dialogues. To achieve this, it is helpful to look at a framework known as the conversation design canvas. This consists of 3 components:
This canvas is then taken and briefed to two stand-in “actors” who will carry out a natural conversation in the form a role play - one playing the bot, and the other playing the user. s
Have a third party observe and take notes while observing this conversation. Note down instances where there were issues and miscommunications in the conversation.
The sample dialogue is then used to create a flow-chart which will list the steps in an average conversation - from understanding the intent of the user to listing all the potential responses or transactions to be carried out and any unforeseen steps that might need to be added before the task can be completed by the bot.
The sample dialogue created in the previous step will most likely be far from perfect. This is where an expert rewrite comes in, where you apply advanced copywriting skills to polish the conversation and make it flow better. There are numerous copywriting techniques that can be applied to do the rewrite but a few critical ones are as follows:
A rewrite like this can also be considered to be a form "internal testing”, where someone within your own team has a look at the effectiveness of the bot conversation in order to make improvements.
Once the sample dialogues and flow charts have been polished by a copywriter it is time to test the conversation design with external users. Get the help of someone outside of the project team to stand in as a new user. Ask them to converse in relevant channel and brief them that they will be interacting with a bot. This could be for example via a WhatsApp business profile, a Facebook Messenger channel or a website’s chatbot interface.
Write down the dialogue created in a Word or Google Document and paste texts from this into the channel in response to the user. Remember to use responses just from the document and don’t improvise or adjust these if there are any hitches. If you observe any gaps or confusion faced by the user, that suggests an area where the conversation design can improve to cover more scenarios.
For a deep dive on testing chatbots, check out our article on the 7 golden rules of testing conversational AI solutions
When you conduct the testing with real users, you will most likely discover scenarios you had not planned for. Consider a bot that helps visitors to an insurance website. The users of the bot could either know exactly which product they want or they may be completely new and looking for guidance. It is important for the bot to recognize the intent in each case and to cover all scenarios.
Responding to each intent involves slightly different conversation design practices.
One approach is to list all the tasks the bot can perform at the start, like this.
Bot: Hi, my name is ALICE. I can help you learn about our products, compare policies, request a quote, submit a claim, talk to your agent or contact our customer service representative. What would you like to do?
While this is informative and covers all the scenarios, the prompt is too long. A more natural approach involves breaking down the scenarios into smaller, shorter prompts to recognize the user’s intent by elimination. It would go something like this very simplified example:
User: Hi, I need help with my insurance
Bot: Sure, would you like to learn more about our products?
Bot: Would you like to compare policies?
User: No I am already your customer
Bot: Perhaps you would like to submit a claim?
Bot: Would you like to talk to your agent?
While this makes the flowchart and conversation longer, it is more natural and easier for the user. Yet another way this can be simplified is by providing buttons or menu items for the user to click and confirm their intents. The same conversation above can be simplified with buttons as follows:
User: Hi, I need help with my insurance
Bot: Sure, what would you like to do?
In most cases, not all conversations with a chatbot will be valuable for the business. There will be a significant proportion of cases where the users involved are many and their value to the business is high (head), and a few situations where the number of users involved are fewer and the value to the business is less as well (long tail). In designing conversations, remember that you don't have to write for every scenario.
The famous Pareto principle, or the 80/20 rule also applies for chatbot conversations. Here it means that 80 percent of your users are going to be using 20 percent of the paths that you design. This is the most important part of the conversation and where the most value lies for your business. The remaining 20 percent is often comprised of edge cases which can be handed over to a human agent, or when the user can be informed about the bot’s limitations.
To find out how KeyReply can help you design engaging and useful chatbots for your organisation, talk to our experts today.