How the Dicobot fine-tuning is going

How would you summarize Dicobot’s beta phase from your perspective?

Kauê Coelho: During the chatbot beta phase, our goal as UI designers was to create an intuitive and engaging user interface that facilitates seamless interactions, emphasizing clarity and ease of navigation, which was instrumental in enhancing the overall user experience. Also, we wanted to humanize the digital interaction by designing a visually appealing and emotionally expressive avatar that could sum up diconium’s DNA in its look and feel. We aimed to create a character that users could connect with on an emotional level. To achieve this, we conducted research on benchmarking other chatbots, facial expressions, body language and different graphic styles.

Diana Andalúz: As for any project, we started by doing research on state-of-the-art chatbots and their UX best practices. We then focused on the main goals for the beta phase. First one was to fine-tune the default messages and responses of the chatbot. Secondly, we wanted to ensure the user feedback was incorporated as both an opportunity for the bot to learn with each conversation, as well as for the user to have an option to provide feedback in a quick and easy way.

Selman Özen: Our legal scope of the beta phase was to keep Dicobot in a secure, internal, and privacy-compliant environment for the time being. We found out that there are several providers for LL.M. models and hostings that are already active in Europe. After an internal assessment of Open AI Azure by IT Security, Data Protection and Legal, we are now ready to integrate this model as well.

What are your overall personal experiences and learnings from the beta phase?

Kauê Coelho: The beta phase was a great learning experience. It provided us insights into user behavior and preferences, emphasizing the importance of user-centric design. Working closely with users and stakeholders allowed us to understand their pain points, which shaped our design decisions. Additionally, collaborating with developers and UX designers and integrating their feedback emphasized the importance of effective cross-functional communication for a successful project. That was also an opportunity to understand the power of visual communication. Designing Dicobot gave us the opportunity to explore the importance of subtlety in expressions further; a slight change in the curve of a smile could convey empathy or reassurance. Additionally, it deepened our understanding of animation principles, enriching our design perspective.

Diana Andalúz: From UX perspective once again it became clear, how complex it is to train a chatbot and define user scenarios due to the broad conversations that it can spark. In collaboration with the team, we learned that a chatbot is like a kid, and it takes time for it to learn, grow and practice its conversational skills.

Selman Özen: Challenges and additional blockers are part of any project work. Our goal is to identify the challenges at an early stage and to consider them preventively already during the individual phases.

What was your biggest challenge in the beta phase and what are your tips to solve this?

Kauê Coelho: Our major challenge was an initial information overload in the interface, causing confusion among users. To overcome this, we simplified the interface, focusing on essential features first and providing contextual guidance. Our tips to avoid this is clear communication, maintaining open channels with the development, UX and stakeholders team to ensure seamless integration of design changes. Regarding Dicobot the challenge was striking the right balance between realism and simplicity in the avatar’s expressions. Our tips:

User Empathy: Put yourself in the user's shoes to understand how different expressions might be interpreted.

Cross-Disciplinary Collaboration: Collaborate closely with UX and UI designers, ensuring a cohesive user-avatar experience.

Iterative Prototyping: Prototype various expressions early and test them with real users to catch emotional resonance.

Diana Andalúz: The biggest UX challenge was to align for expectations on the level of accuracy that this initial version of Dicobot would have. Tips to solve this would be to have frequent alignments with the experts involved in the development of the AI solution, for us this was really insightful to better understand how a chatbot is created. In order to align on expectations within and outside the team, another tip is to define a prioritization and progressive rollout of the optimal and end-goal experience.

Selman Özen: Our appeal to all (data) product teams is quite clear: let the lawyers and data protection experts be a part of the project work early on.

In your unit, what are the next steps in the development of Dicobot?

Kauê Coelho: For UI there are lots of possibilities to explore including different viewports, integration with the website and merchandising. Our plan involves enabling the avatar to respond to user using various expressions depending on the content. This means the avatar can adapt its expressions dynamically, creating a more personalized and empathetic interaction.

Diana Andalúz: Next steps would be to work together with development and AI experts on prompts that will support the most common topics and to expand user feedback so that Dicobot can learn even more from each conversation. We will enhance Dicobot’s personality and tone of voice. You can also expect further development on the different viewports and ways to incorporate Dicobot into our company website.

Selman Özen: With the help of determinable business cases and roll outs, we will now check whether the requirements from data protection law – also in relation to the planned EU AI Act – are met.