3D Cloud
Ai Assistant
Project Type: New B2C Product Development
Primary Tools Used: Figma, Jira, Slack, Playbook UX
Overview
Due to the method employed by 3D Cloud in creating their 3D assets, each product is generated with multiple mount points. These points are assigned specific references that define what can and cannot be attached to each point, with most references being universal. As a result, clients like MillerKnoll, who maintain a database of over 30,000 3D assets, have access to more than 1,000,000 unique possible combinations.
The task
Create an AI assistant that collects details about the client's needs, lifestyle, preferences, and budget, then uses this information to narrow down the choices to the most suitable options.
My Approach
AI was still an emerging technology when this project began, and there were numerous uncertainties that required thorough research, as well as growing concerns about how AI integration would impact daily life. With this in mind, when we decided to explore the AI technology market, we opted to begin with a simple question and answer style interface. We were transparent with users, ensuring they knew they were interacting with AI, and provided them the option to opt out at any point, allowing them to continue building their product in the traditional manner.
We began by mapping out potential user flows, identifying key areas where the AI would need to learn. We then cross-referenced these flows with the existing client database, and in collaboration with the development team, we began to feed the AI a set of test scenarios that would enable it to learn and improve its ability to select suitable outcomes. Initially, the AI produced erratic results, with an imbalance between style and functionality. This led to product designs that were either not aesthetically pleasing or practical. However, as we continued to provide the AI with more data, we observed a significant improvement in its results.
With a clearer understanding of what the AI could deliver, we applied a simple UI overlay and proceeded to user testing.
The test revealed that while the AI tool effectively narrowed down the options for users, they were hesitant to trust the results and often sought ways to clear the suggestions in favor of browsing the full dataset. Additionally, feedback indicated that users were uncomfortable with the AI asking specific questions regarding their home and health.
In response to this feedback, we decided to integrate the AI with the Google search database, enabling it to continuously learn and leverage additional data such as user reviews, sales results, and social design trends. We also redesigned the interface to adopt a more straightforward, customer service-style chat format. This adjustment allowed the AI to pose the same questions as before, but with users feeling that they guided the interaction, which in turn made them more comfortable providing responses.
With a fully functional AI chat assistant that had been tested and proven effective, the sales team began marketing it to our existing clients. My team was then tasked with customizing the user interface to align with each client’s specific brand requirements.
