Learnings from designing conversational UIs for a year

During my time at Springbok agency, I was the UX/UI specialist for the Conversational User Interfaces Team, a multidisciplinary squad of UX researchers, strategists, conversational architects and software engineers with one goal: transform the hype on chatbots to real amazing experiences.

The challenge

Customer service and the overall pre and post-purchase experience are becoming more and more critical, and if we wanted to harness the hype, there was no time to waste. Acknowledging the limitations, from click-bots to AI-powered solutions, was vital for the engineering team behind our proposals and me as a designer. How could I create experiences that were not upsetting? That was human? Our clients were among the most prominent enterprises in the Benelux area, and they approached us to challenge their current CUI solutions and improve on them.

Our outcome had to be technically feasible and compliant to regulations, but it also had to fix actual pain points, we didn't want to launch useless obstructive pop-ups taking up screen real estate. Add maintaining brand coherence and, more especially, tone of voice to the mix, and we had ourselves a challenge. We worked with Proximus, ING, Engie, and AXA, in proposals and shipped products, and here are some of my contributions and learnings.

My role

My tasks were to transform the insights we gathered through our experience, user testing, and our assumptions, crazy, almost sci-fi wishes too, into actionable prototypes that we could pitch, iterate on, develop, and launch.

Case #1: Engie's FAQ Bot

Goal: Engie is one of the leading energy companies in the Benelux. The company's request was to help them reduce the amount of inbound calls to their call center. We went to the root of the problem and partnered up with the customer satisfaction team. Most of the questions asked were simple and had visible answers on the FAQ page. But this page was not optimized. One design team was in charge of quick wins, while the CUI team took the opportunity for innovation.

Execution: Placing a chatbot ASAP to start testing and training was the first solution, but it was not enough: we knew the answers needed to be more visual and intuitive. Not to mention accurate. After tests, competitive analysis and research on customer satisfaction surveys and online comments, the results showed that most users couldn't find straightforward answers to their specific situations.

• This chatbot needs to be in a context-relevant space, not everywhere on the website. Its purpose must be specific.
• It has to be direct and clear on what it can or can not answer (this is part of our CUI bible)
• It had to go the extra mile: Need instructions? Here's a video.
• Chatbots could be new to the typical caller. A friendly but professional tone of voice and clear instructions were needed as soon as the engagement started.
• We don't want to hide information from the user, so our UI would not be obstructive.

Case #2: ING in app

Goal: The ING bank approached the agency and we were challenged to look for pain points on their current assistant (available on Facebook Messenger) and show our views on the future of AI assistants.

We analyzed the company's current bot during an intense workshop and found problems in tone of voice, usability, and a mismatch between the bot's offer and the user's mental models. We developed an in-app bot prototype that can make it seamless to access help on the bank app.

Case #3: AXA contextual and conversational questionnaire bot

Problem: We were asked to create a critique document and suggest solutions for the poor engagement rates of the current bot on the AXA website. The underpinning goal was to help reduce the bounce rate on quotation forms and improve self service.

We foresaw a short and long-term fix. In the short term, we could train the existent bot and create visual milestones on the conversation so users can receive answers more coherently. The long-term solution was to address the conversion and bounce rate problem with conversational cues, improved tone of voice, and chunking information in relevant packets.