ChatGPT based Experience for UCL Final
Adidas and UCL partnered to create a in-game experience for the Champions League final. Using ChatGPT as base for this experience, users could interact with a persona via a WhatsApp chat to talk about the match and receive exclusive Adidas drops during the game.
The Ask
Adidas and UCL joined forces to deliver an immersive in-game experience for soccer fans belonging to Generation Z. The innovative concept involved a WhatsApp chat interface that allowed users to engage in a conversation with a GenZ idol, discussing pivotal moments of the game. Users could ask questions about the Champions League and receive exclusive Adidas drops that were exclusively available during the game. This collaboration aimed to captivate and engage young soccer enthusiasts, offering a unique and interactive way to connect with their favorite sport and brand.
Project Overview
What defined success to my UX Work
Concept Validation
To facilitate informed technical decisions, we opted to conduct a Concept Validation. For this purpose, we applied the Wizard of Oz research method, where a simulated experience is created for users, even though the underlying technology may not yet be fully functional. We were able to gather valuable insights and feedback from users while still in the development stage.
Technical Constrains
The project encountered significant technical constraints. With the development occurring during the quarter-finals, there was limited time available to implement highly intricate features. Therefore, the focus was on ensuring the fundamental aspects were executed flawlessly to deliver a remarkable user experience.
UX to Inform Key Decisions
UX played a pivotal role in informing key decisions . In this context, the UX team was tasked with creating Jira tickets for development features and stories, as well as constructing a User Journey. These deliverables served as crucial tools to guide and inform technical decisions throughout the project and align expectations with internal and external stakeholders.
Concept Testing
Helping inform creative decisions
In order to explore this concept, we decided to conduct a Wizard of Oz (WoZ) Concept Testing using an actual Premier League match between Everton and Tottenham Spurs as a reference. We divided the participants into two groups, comprising a total of 14 individuals with diverse backgrounds, including soccer professionals and non-soccer fans.
During the first half of the match, we conducted an unmoderated, closed paradigm test. In this test, all responses were pre-scripted before the match started and could not be altered. For the second half, we implemented a moderated, open paradigm testing approach. We generated live responses using various ChatGPT prompts. The participants were allowed to engage in real-time conversations, and their responses were not scripted in advance.
Throughout both tests, the underlying logic remained consistent: users needed to mention "@username" to prompt a response from the system. To specifically assess the impersonation aspect of ChatGPT, we adopted the writing and speech style of the British rapper Stormzy.
Sensitive information regarding participants' identity were carefully erased from this screenshot.
Key Experience Findings
Analysing data from the Concept Testing
Prior to conducting the Concept Testing, our team had several concerns regarding the dynamics of game interactions between the persona and the users. We wanted to address these questions in order to ensure a smooth and enjoyable experience for everyone involved.
One of the primary concerns was related to the text messaging flow. We wanted to ensure that the chat system would be able to handle all incoming messages effectively. Additionally, we questioned whether users would understand the basic structure of the chat and how it functions.
Another significant aspect we focused on was the quality of responses generated by the chat. We wanted to avoid any potential issues of racist or biased answers from the system. Furthermore, we needed to determine how the chat would respond if someone expressed racist views during the match. It was crucial for us to create a safe, inclusive, and engaging environment for all participants.
Following the completion of the test, we gained valuable insights and learned important aspects:
- 90% of users understood basic commands and instruction;
- The bot wouldn't be able to understand any media sent (eg. images or audios) and therefore we needed to disable that functionality;
- ChatGPT know how to moderate conversations very well when responding to harmful answers;
- ChatGPT was able to incorporate knowledge from previous responses (eg. score and red/yellow card informations).
Some attention points were:
- In longer session, ChatGPT becomes erratic and prompts need to be regulated again;
- During the testing, we observed that ChatGPT was capable of impersonating convincingly, adopting the writing and speech style of British rapper Stormzy as intended. However, there were instances where ChatGPT deviated from the desired impersonation;
- Because of the large number of messages during testing, we decided to impose a limit on the total number of messages that could be sent per minute in the real experience. This limit was put in place to ensure smooth interaction flow and manage the system's workload effectively.
User Flow
Helping informing important decisions
To inform creative and technical decisions and facilitate effective communication with the client, we developed a comprehensive user flow that depicted all user interactions and touchpoints. Additionally, we mapped out potential opportunities to send Adidas drops and engage users with conversation starters. Furthermore, we included examples of messages to provide a clearer understanding of the user experience.
We have also conducted user interviews with the concept testing participants, to gather insights about the most desirable features, understand user expectations and receive direct feedback of the experience.
What I've learned?
This project stands as my favorite of all time. Throughout its course, I gained extensive knowledge about ChatGPT technology, honing my skills in crafting effective prompts and navigating its intricacies. Additionally, I delved into the realm of data analysis and UX research, with a particular focus on the Wizard of Oz method. Conducting follow-up user interviews further expanded my understanding and allowed for deeper insights to be gleaned. The experience has been invaluable in advancing my expertise in these areas.
