What’s the key to a successful chatbot project?
This article is based on a presentation first held at Finance Norway and ICT Norway’s Fintech conference, FutureBank 2019.
Chatbots are experiencing rapid growth. An increasing amount of companies offer automated chats on their webpages, and Facebook Messenger now hosts more than 300 000 chatbots. But there is a wide gap in both the technology and the efforts behind these chatbots, and that can cause skepticism. Do we actually need chatbots? What if they don’t give the right answers? Because even if 300 000 might sound like a lot, chatbots are still in a starting phase. To predict the future of chatbots, we first have to look at what works well today.
So, what is the key to a successful chatbot project?
Businesses enjoy several perks when implementing chatbots. Chatbots are always available and respond immediately, regardless of the hour. This creates shorter waiting lines, meaning a shorter waiting time for the end user. A chatbot is not supposed to answer everything, but it can answer the most frequently asked questions, which again frees time for human customer service agents to focus on the more complex inquiries.
But all of this requires that the customer can trust that the chatbot performs well.
We at Convertelligence have created our own chatbot platform, Kindly, where our customers can build and tailor their own chatbot. Kindly combines strong machine learning engines with a user-friendly interface. Using API integrations, Kindly can also be connected to different systems, which allows the end user to perform actions directly through the chatbot. For instance, some of our chatbots have the ability to book a hotel room through the chatbot, scan goods in a store or offer a loan calculator.
But good technology and a user-friendly platform is not everything. When a chatbot interprets a question correctly and gives the correct answer, this is because the chatbot has understood the intention of the question. For a machine to understand an intention, it needs large amounts of varied and qualitative training data.
Therefore, UX writers from Convertelligence work closely with our customers throughout the chatbot project, to ensure the chatbot gets a sturdy foundation of data. Building a chatbot can sometimes feel repetitive and time-consuming, but going the extra mile will be worth it in the end.
So, what does a chatbot project consist of?
In Convertelligence, we divide the chatbot project into four main parts: preparatory analyses, training, building period, and handover.
Preparations are important before embarking on a chatbot project. Convertelligence, therefore, offers both quantitative and qualitative analyses to map out what the customer expects of the chatbot, and what it should be able to answer. This makes it easier to define the scope — or focus area — of the chatbot and to know how end users phrase their questions.
After the analyses, Convertelligence offers thorough chatbot training for all the customer’s future chatbot builders, where we give an introduction to Kindly and chatbot building in general.
Here, before beginning the building period, the customer needs to consider some important questions: Who, what and how?
Who is going to use the chatbot? Is this an external chatbot or maybe an internal one, perhaps aiming to help with the onboarding of new employees?
What does the chatbot need to know? What kinds of questions should the chatbot be able to answer, and what does it not need to answer? What’s the chatbot scope?
And, how do you want to build the chatbot? Do you want to build the chatbot from scratch or do you want to implement premade content, so-called skills? (See our article Using a chatbot template: How pre-made content drastically cuts chatbot assembly time for more information about the benefits of skills).
It’s also wise to consider from the onset how the chatbot should phrase its answers. If you contact your bank, asking for a loan, would you be happy with this response?
No? Why not? Cash bot does promise you money. But perhaps this answer doesn’t match your expectations of how a bank should respond. And if the answer doesn’t fit, can you trust it?
Different companies will have different voices, and thus their chatbots should speak accordingly. The chatbot is an elongation of the brand. Creating a design guide before embarking on building ensures that all chatbot builders use the same tone of voice when writing the answers.
Making all of these decisions beforehand will save time during the project, preventing the chatbot builders from having to go back and make changes at a later stage.
Building the chatbot
During the building period, UX writers from Convertelligence visit our customers, and together, we build and test the chatbot foundation. The most important part of this period is knowledge transfer. The customer has the expertise in their company and Convertelligence has the expertise in chatbot building. Together we build the foundation of the company’s automated chat.
It is essential that the customer’s chatbot builders feel confident about chatbot building at the end of the project, to make them comfortable with handling the chatbot on their own after the handover meeting.
Because after the project period, the customer continues to build and handle the chatbot alone.
This does not mean that Convertelligence loses interest in how the chatbot performs. On the contrary, chatbot performance is very important to us, and we continue to offer chatbot analyses and follow-up meetings to ensure the chatbot provides the right answers. We also have regular user forums where our customers can exchange chatbot experiences and learn about our new features and upcoming plans for Kindly. We want to ensure that our customers can trust the chatbots to perform well.
So what is the key to a successful chatbot project?
Convertelligence recommends preparing thoroughly, selecting a clearly defined scope, making important decisions — such as decisions regarding the tone of voice — early on and focusing on competence transfer during the project period.
Strong machine learning engines and intuitive chatbot tools are important, but a chatbot is nothing without the people building it.
The key to a successful chatbot project is, therefore, to ensure that our customers learn enough about Kindly and chatbot building during the project to be confident about maintaining and managing their chatbot on their own afterward. Kindly is artificial intelligence — activated by humans.