Inside Just Eat's AI Lab: Voice Agents & Agentic Commerce

The podcast discusses Just Eat Takeaways evolution into a multifaceted platform offering food delivery, grocery, electronics, and other services, emphasizing innovation to meet consumer demands for instant convenience. The companys incubator operates with startup-like agility, testing emerging AI and developing prototypes in 24 weeks. Key projects include voice-based features with conversational capabilities, context-aware recommendations, and integration with chat and search platforms. The firm explores future interaction models like "super apps" or no-app scenarios, aiming to adapt to shifting consumer behaviors, such as the 60% year-over-year growth in chat-based discovery. It also investigates using wearable data for personalized recommendations, like suggesting post-workout meals, and addresses challenges in seamless voice-to-action integration and cross-platform context retention.

The podcast highlights the importance of balancing technological innovation with user-centric design, focusing on personalization that aligns with customer preferences, behaviors, and health goals. Challenges include ensuring voice agents reliably execute tasks (e.g., opening apps, modifying orders) and overcoming ecosystem fragmentation by collaborating with tech companies and adhering to data privacy regulations. The company prioritizes avoiding over-servicing through irrelevant notifications and refining AI agents to handle complex requests, such as event planning or dietary-specific recommendations. Additionally, it explores agentic AI commerce, integrating AI across voice, chat, wearables, and logistics (e.g., drones, robotics), while addressing friction points like user consent, intent preservation across devices, and the ethical use of data. The discussion underscores a commitment to iterative testing, problem-solving, and aligning AI advancements with practical, user-friendly applications.

• What if you prototype a voice-based AI agent that integrates with wearable health data to suggest personalized meal options in real-time?
  Move: Build a prototype using a wearable API (e.g., Fitbit or Apple Watch) to pull health metrics (e.g., heart rate, activity level) and feed them into a voice agent that recommends meals, adjusts for dietary restrictions, or suggests post-workout snacks.
  Why now: The text highlights the growing use of wearables for health data and the importance of contextual recommendations. Voice interfaces are also rising in preference, aligning with the companys focus on conversational continuity.
  Expected upside: Faster user adoption of hyper-personalized services, reduced friction in ordering, and a competitive edge in food/delivery markets.

• What if you test a no-app chatbot model that uses existing messaging platforms (e.g., WhatsApp, Discord) to handle full commerce flows, like grocery orders or electronics replacements?
  Move: Develop a chatbot that runs on popular messaging apps, leveraging their built-in user bases to avoid the need for a standalone app. Integrate voice-to-text or chat-based interactions for seamless ordering.
  Why now: The text emphasizes the shift to "no-app" models and the 60% YoY growth in chat-based discovery. Using existing platforms reduces development overhead and aligns with consumer trends.
  Expected upside: Lower user acquisition costs, faster scalability, and alignment with future consumer expectations for seamless, app-free interactions.

• What if you build a voice-to-action system that modifies orders in real-time based on context-switching (e.g., voice commands while biking, without touching a device)?
  Move: Create a voice agent that maintains context across devices (e.g., starting a conversation on a smartwatch, then continuing it on a phone or smart speaker) to allow users to adjust orders (e.g., Cancel the pepperoni, add vegan cheese) while on the move.
  Why now: The text notes challenges in voice agent reliability and the need for context-aware intent. This directly addresses the friction points in voice ecosystems and the companys focus on agentic AI commerce.
  Expected upside: Enhanced user convenience, reduced cart abandonment, and a stronger position in the race for seamless, hands-free commerce experiences.

• Build MVPs Rapidly with AI Prototypes: Use the Preseed teams 2-4 week timeline model to develop AI-driven features like voice agents or chat commerce interfaces, focusing on quick iterations and user feedback.
• Leverage Wearable Integration for Personalization: Incorporate data from fitness trackers or health devices (e.g., heart rate, activity levels) to tailor recommendations (e.g., suggesting protein meals post-workout) and create proactive user prompts.
• Ensure Cross-Platform Context Retention: Design systems to maintain conversation flow and user intent across devices (e.g., from voice to mobile), while respecting user consent for data sharing between platforms like ChatGPT and your app.
• Test Voice-to-Action Integration: Address common voice agent friction points (e.g., Siri failing to open specific apps) by rigorously testing seamless voice command workflows (e.g., modifying orders via voice while on a bike).
• Prioritize Privacy-First Data Practices: Implement strict user consent mechanisms and transparent data sharing policies when integrating across platforms (e.g., OpenAI, wearables) to comply with regional regulations and avoid consumer mistrust.