Chatbot conference (Utrecht)
Author
Maarten Vos
Published
15 June 2018
Reading time
4 minutes
At the end of April I attended the first Chatbot Conference in Utrecht. An event about chatbots, conversational copy and virtual agents. I attended three presentations.
Presentation 1: Investing in Natural Language Generation
Hans van Dam (RoboCopy) explained why conversational copywriting is so important. Users want an experience that is as natural as possible, as if they were talking to friends. Chatbots therefore need to be helpful, human and able to hold a persuasive conversation. We want chatbots with personality and empathy, who are helpful, have natural language and are also persuasive. And that's where conversational copywriting comes in. Van Dam distinguishes various elements in this regard.
- Character development: Think of brand identity, tone-of-voice or the uncanny valley.
- Scenario development: Intent – obstacles – resolution.
- Conversation design
- UX writing (“Don't make me think”)
- Service copywriting: Service-oriented dialogues ("What do you need help with?", "I'm going to try to help you find a solution.") and if it doesn't work, refer to the customer service staff.
- Escalation Prevention: To prevent complaints, revenge and termination of the relationship by the customer.
- Persuasive copywriting.
- Data driven optimization with data that can adjust the dialogues.
- Technology & tools.
How do you teach the user to use the chatbot? Tell him to type short questions. You can also achieve this with input fields that allow a limited number of characters. Have the bot give answers like "I'm still learning".
Presentation 2: A look behind the scenes at a chatbot selection and implementation
Muriël Surrurier Schepper (appliedai.nl took us through the process of selecting and implementing a chatbot, in this case for a large bank. This process consisted of nine steps.
- Internal kick-off: The main question was why do we want a chatbot? Answers can include lower costs, more consistency in answers, increase customer satisfaction and 24×7 availability.
- Request for Information & Requirements: These were requested from a mix of providers: major IT players, specialist chatbot suppliers and start-ups. No fewer than 180 requiremants were drawn up in the form of user stories: “As a [type of user], I want [some goal] so that [some reason].”
- Demonstrations: Ultimately, five suppliers were invited. The availability of a Dutch language model and production experience was important.
- Request for Proposal and reference visits: Three suppliers received an RfP and customers of the suppliers were visited. The customer stories turned out to be very valuable. Two suppliers remained.
- Contracting: Suppliers worked under equal conditions, in terms of budget, scope, KPIs, chat history, time and available hours.
- Chatbot training: Both vendors got to work populating their bot. One onsite, the other offsite. One waterfall, the other agile. One with a focus on content, the other on IT. One involved content agents closely from the start, the other put their own specialists to work. One tested extensively with the agents, the other internally. In the end, onsite, waterfall, content, and closely involving agents and testing with them turned out to be the most successful.
- Production pilots: After (partial) filling of the bots, the two production environments went live at the customer. Employees were asked to test the two bots.
- Experiences: People seem to like testing the bot by saying weird and funny things. Answers are not always trusted and users like to have an 'escape' to a live chat. Similar chatbot outcomes lead to different ratings. Positive experiences: We should have done this much sooner; Friendly, fast, modern, easy, clear, helpful; I'd rather be a bot than talk to someone on the phone. Negative experiences: Irritating and impersonal; Worthless, no sensible answer came out; Sad that personal attention is not available anymore.
- Lessons learned: It was a great experience to have two chatbots in competition with each other. The pilots provided good insight into how to train and work with a bot. Not all customers will be happy with a bot, so think carefully about the scope and implementation strategy. A bot is and remains a lot of human work. Reporting and tooling are very important in order to continuously optimize the answers. Writing those answers is something you have to learn.
Presentation 3: Why a.s.r. built a conversational website
Nick Aanraad (Lead Marketing & Digital at a.s.r.) talked about the 'first fully conversation-oriented website in the Netherlands'. From one day to the next, a.s.r. switched from a 'normal' website to a conversational interface. In fact, the entire website is one big chatbot. Why this huge change?
The old a.s.r. website no longer met the needs and expectations of today's user. Consumers are focused on looking for something and do so via Google or comparison sites. And customers who did not achieve their goals on the website called customer service. That is why a.s.r. wanted to help the existing customer better, faster, more direct and more digitally via the website. The customer's question had to be the focus, instead of a huge supply of information that the customer might find useful. In addition, a.s.r. wanted to reduce e-mail and customer contact by telephone and increase online customer contact. Three agile teams built the completely new website in just over six months.
And then the site went live. Initially, there was great dissatisfaction. The response on social media was very negative and the press wrote about the issue very negatively. But a.s.r. persisted and immediately started collecting data. Customers turned out to formulate questions slightly differently than expected and they responded immediately.
An important change was the introduction of 'conversation starters'. Initially there was only an input field with 'Ask your question'. Now customers can choose from five topics: Close a product, Report damage, Change data, Funds & prices and Contact, still followed by the input field 'Ask your question'. This turned out to be much more successful than an 'open' start.
Also, 'Digital' now has faster insight than customer service when something is not clear to the customer and can respond immediately. For example, prevention tips and proactive content are shown to relevant visitors at relevant times. The number of successful dialogues has increased enormously due to many content optimizations.
The development of the site is ongoing. For example, the wish list includes recognizing the customer across all business units, serving the customer personally in a logged-in environment and increasing the intelligence of the bot. All in all a very interesting day. The development of chatbots, both technically and in terms of content, is in full swing. For many content professionals, this is an unexplored area where many opportunities lie. Many companies will need help in the (near) future in setting up and maintaining their chatbot as efficiently as possible. So plenty of options.
About the author

CUIs
Events
Chatbot
Events
User experience
UX