Inspect & Upsell

Inspiration

We wanted to mimic a real salesperson as closely as possible in an automated environment. We asked ourselves why people consistently skip add-ons during online purchases, like when booking a flight, while they often accept the same offers when a human presents them. The difference is the human interaction: rapport, trust, and subtle persuasive cues. Our goal was to bring these elements back into the automated key-collection flow.

What it does

The system keeps the interaction as human as possible through a conversational avatar. It uses all available customer information—trip details, budget, preferences—to propose accurate upgrade and protection suggestions. It applies known sales techniques to gently persuade the customer to spend more, recreating the natural “sales moment” normally performed by SIXT staff.

How we built it

We built the backend in Python. We used a large language model to manage the conversational flow and persuasion logic, integrated a D-ID avatar for a realistic human interface, designed a lightweight UI to collect customer needs and budget, and created a targeting mechanism that aims at offering options around 130% of the declared customer budget.

Challenges we ran into

We had to recreate a natural conversational flow without making the agent sound scripted. Keeping the entire interaction live with very low latency was technically challenging. It was also difficult to balance persuasive upselling with a pleasant customer experience. Integrating the UI, backend, avatar, and LLM smoothly required significant effort.

Accomplishments that we're proud of

We built a functioning human-like digital sales agent within the short hackathon timeframe. We created dynamic conversations that adapt to user input, implemented a realistic upselling strategy, and designed a solution that can be integrated into the existing automated key-collection process.

What we learned

We learned how much subtle human cues matter in sales and how to guide an LLM to use persuasive techniques responsibly. We saw how avatar-based interaction increases engagement compared to a flat UI. We also learned how to balance psychology, UX design, and engineering to create a believable sales interaction.