This project was developed for the Audi Innovation Award Hackathon 2023, with the aim of transforming the Audi car-buying experience in Taiwan. By utilizing sentiment analysis, this innovative AI sales agent streamlines purchases, offers personalized guidance, and expertly negotiates, making Audi more accessible and appealing to Taiwanese customers.
MY ROLE
User Research, Define, Ideation, Mockups, Prototyping
COLLABORATED WITH
Data Analyst/Backend *1
Compliance RAQA*1
Product Manager*1
DURATION
• Hackerthon : 3 days, full time, August 2023
• Refined design : 1 month, September 2023
Problem Space
Despite being a premium brand, Audi faces stiff competition. According to the 2022 Taiwan Car Sales Ranking Report, whether from the overall market perspective or the imported luxury brands, Audi always finds itself trailing far behind. Audi's main strategy has been trying to focus on enhancing digital experiences with the aim of improving customer satisfaction.
It has become evident that new potential customers often face challenges and uncertainties after we talked with Audi sales agents and gathered information via web crawler, including:
The impact
By gathering data that traditional sales systems couldn't achieve before, it provides more precise and efficient marketing methods, whether for a chatbot or a sales team. It also reduces customer hesitation and increases the likelihood of closing a deal.
Limits
The compressed timeframe of the 3-day Hackathon restricted our capacity to develop a fully realized prototype. Consequently, we focused on the business strategy and technical components for the final presentation. We encountered obstacles due to limitations in revealing confidential information supplied by Audi. Following the event, I dedicated an additional 3 weeks to delve deeper into the project and finalize the detailed design.
Talk with Sales Agents
Before delving into the project, we engaged with Audi's sales agents to gain valuable insights.
They highlighted the frustrations experienced by both customers & agents when dealing with complex forms. This set the stage for our design challenges.
Customers who visit the showroom often hesitate to provide all the required details on the forms, making it challenging for the sales team to collect customer data for future marketing purposes.
Why Do Customers Hesitate?
Researches show that there are 5 major factors related to consumer behavior.
Resources from Encyclopedia of Psychology, American Psychological Association.
While it's important to note that individual reactions can vary widely based on personality, past experiences, and situational factors, the exact nature of hesitancy and anxiety can differ from person to person.
Identifying the Pain Points
We now know that word of mouth is one of the essential factors that affects customer's decisions.
After conducting research via web crawler, we found that the majority of the reviews, whether from online forums or customer feedback data, are negative.
But how can we advance the consideration process?
Brainstorming
We attended a Value Proposition workshop. Using this model, we analyzed if Audi's current products and services fulfill the customer's jobs, desires, and relieve their pains.
We identified what can be resolved first at consideration stage from the right side, and then developed updated solutions on the left.
Early Ideation
A streamlined, AI-powered online car buying process was our goal.
At the early stages, we've explored multiple solutions. Our early ideation aimed to develop a simplified buying process, which involved the following 3 key steps:
However, things don't always go as planned
Our initial solution was overly ambitious, and required a refocus.
Despite our initial direction, we faced technical challenges. After consulting with professionals, including a senior PM, a UX designer, and the Head of Strategy, we realized that it was not feasible to create a comprehensive business plan with all the intended features within the given timeframe. Therefore, we had to make a difficult decision before the night of presentation.
With a last-minute pivot, we decided to concentrate on the AI sales aspect.
Users visiting the official website could access the chat feature, enabling direct interaction with the AI sales agent.
Why we did this, even though Audi Taiwan already has an official LINE chatbot?
Audi Taiwan established an official LINE account in 2020, recognizing its popularity as a communication tool in Taiwan. However, there is room for improvement in certain areas.
So what's so special about AI Audi Navigator compared to a regular chatbot?
Clustering Car Models by Personality Tag
The AI chatbot employed Support Vector Machine (SVM) technology, utilizing a dataset of around 15,000 pieces of Voice of Customer (VOC) data for personality tagging.
This allowed us to categorize customers into different groups based on their preferences and personalities.
For example, we've identified 2 main groups for A-series car owners vs Q-series car owners from the database we had during the Hackerthon.
4 Negotiation Modes
We established these 4 negotiation modes based on sales team feedback. By utilizing LLaMa, allowing the AI chatbot to tailor their interactions based on the user's personality traits, ensuring the most personalized and effective service.
How it works
The AI collects chat records and performs data analysis, with the goal of helping the Sales team create customer personas, ultimately preparing them to provide better customer service.
Users can ask questions in natural language regarding personal guidance, accessory combinations, and even pricing. If users prefer a human representative, they have that option too.
Set Up
Before establishing conversation flow, I needed to determine the conversation elements for the chatbot.
Elements
Persistent Menu/Quick Drawer
Minimum Viable Information Architecture
I mapped out scenario use cases to create seamless conversation flows.
Hello : Triggered once a user clicks on the "Get Started" button. This will be the default entry flow.
Catch All : Triggered once a user does or says something that bot doesn't understand.
Human Handoff: Triggered when they want to talk to a human.
Goodbye: Triggered once a user completes the experience.
Use Cases and Data Gained
There are 3 main tasks taking A-Series car potential buyer as examples of conversation flow and the data that could be collected during each task:
Task 1 : Consult
Task 2 : Request an online quote
Task 3 : Request a test drive
My Reflections
Contact