Just Eat shows how building on strong foundations lets you do both the glamorous innovation work and the high-value automation work, unlocking the real promise of AI on multiple fronts at once.

By picking targeted areas and scaling successes to a shared platform, Just Eat focused on ensuring key systems were available for integration with all flavours of AI agents and workflows. From Customer Service to partner-side processes, and internal tooling that lets colleagues build what they need on the shared platform, available to all roles and functions. By empowering users from finance, legal and other areas to build, they no longer have to wait at the end of an engineering backlog.

That groundwork is what made a consumer voice assistant, to help customers in the discovery phase, a reasonable next step. James Osborn, Senior Director of AI & Data at Just Eat, traces the line from back-office automation to voice ordering running across European markets, explaining what each layer had to be capable of before the next one was attempted. He closes on what the internal work gave Just Eat Takeaway that a standalone voice project would not: employees who already knew how to build with these tools.

Key takeaways

• Which internal and partner-side automation needs to be working before a customer-facing agent is realistic
• How a shared AI platform lets non-engineering teams solve problems engineers were never going to reach
• What Just Eat gained by sequencing capability instead of starting with the visible thing

---

AI Agents are moving from pilot projects to daily operations, but most CX teams are still unsure where the line between hype and real capability sits. This session cuts through the noise to unpack what's actually working: which use cases are ready for autonomous handling today, which still need a human in the loop, and how leading teams build the trust and control needed to scale AI safely.

This is a practical look at real deployment decisions, based on LuxCreo's case study: pairing deterministic guardrails with predictive reasoning so AI Agents follow policy while still handling nuance, using unified customer data to turn a scripted bot into an agent that can actually resolve issues, and measuring ROI beyond deflection rate alone. Attendees will leave with a framework for evaluating their own AI Agent readiness and a clear-eyed view of where the technology actually sits today.

Attendees will learn:

• How to identify which CX workflows are ready for autonomous AI Agent handling today, and which still need a human in the loop
• Concrete examples of AI Agents resolving issues end-to-end (processing a return, updating an order, issuing refunds) without a human needing to step in
• How to measure AI Agent ROI beyond deflection rate, using real numbers from a live production deployment

---

---

---

Consumers measure every brand against the best experience they have ever had anywhere. Their expectations continue to rise, shaped by the AI they use in their daily lives: AI that remembers them, understands their history, and finishes the job in one pass. 

Organizations are adopting AI at a rapid pace, adding more automation, more copilots, more point solutions. Yet the gap between what customers expect and what systems actually deliver keeps widening. Most organizations are not short on data or tools. They are short on contextual intelligence, the ability to bring scattered context together in the moment and act on it.

The problem isn't that organizations aren't deploying AI. The main problem is the organization's operating model – how work actually gets done. While organizations have evolved to an AI-assisted model, with automation handling routine tasks, people remain central to every decision. The organization's real context, such as purchase history, open cases and prior interactions, stays scattered across disconnected systems, unable to be activated in the moment. 

The answer? An agentic operating model, one where AI, people, data and systems are coordinated in real time around outcomes rather than managed as isolated tools. Attendees will see how connecting and compounding context from every interaction changes what is possible, understand the technical and organizational steps to get there, and walk away confident about success in the agentic era.

Attendees will learn:

• Why the AI-assisted operating model, where automation and copilots make existing processes more efficient, hits a ceiling, and what distinguishes it from a true agentic operating model built around outcomes.
• How activating the context an organization already has, across systems, channels and customer history, turns every interaction into intelligence that improves the next one.
• How a unified CX platform that can access context across the entirety of the customer journey is key to successfully adopting the agentic operating model. 

---

---

---

• What are the four levels of autonomy and what can each deliver for CX?
• The questions to ask vendors during demos to ascertain if the product is genuinely agentic or a rebranded chatbot.
• How to recognize use cases for AI agents and know when a more traditional workflow is more appropriate 

---

---

• What a recovery layer looks like in practice, and how it catches failures before customers report them
• Which three roles every AI journey needs named (process owner, data owner and recovery owner) and what each one answers for
• How to score the journeys you already have live, and why recovery quality says more about customer trust than containment rate does

---

Agentic AI is arriving faster than consumers can name it. According to the April 2026 Agentic AI Usage and Perception study, while almost every North American recognises AI and more than a third already use it, 60% have never heard of agentic AI and just 3% have used it.

Drawing on that research, Michael Nevski, former Director, Global Insights at Visa, unpacks the gap between curiosity and usage. Consumers hold hope and hesitation at once, with a majority believing AI will improve their lives but also holding concerns around it. That reluctance appears to be negotiable, with interest in AI shopping assistants increasing significantly.

The session will reveal and analyse the results of the major study, led by Nevski, and closes with a practical view of which customers adopt first, which safeguards move the needle, and where accountability lies when an AI agent makes a mistake.

Attendees will learn:

• A data-backed picture of agentic AI awareness, sentiment, and intent across generations and income levels
• The specific categories and tasks consumers will hand to an agent first - and the ones they won't
• Design and disclosure choices that convert hesitation into adoption

---