Agentic Commerce Unlocked: AI Use Cases Redefining Travel

This flagship session on Agentic Commerce at ITB Berlin 2026 brings together three senior leaders — Piero Sierra (Chief AI Officer, Skyscanner), Sergio Golia (VP Strategy, Travel Unit, Amadeus), and Anne Pruvot (CEO, SNCF Connect & Tech) — to demonstrate real, working AI deployments across meta search, GDS/B2B travel tech, and national rail distribution.

Piero Sierra opens by grounding the session in Skyscanner's scale: 160 million monthly active users, representing approximately 23% of all flight searches globally, with leadership positions across EMEA, APAC, and Latin America. He frames AI as 'the next big thing since the internet and mobile phones' for travel, but critically distinguishes four maturity levels rather than treating AI as a monolith. Level 1 ('Base Camp') is AI sprinkled on existing experiences — for example, using AI to reverse-engineer why a machine learning ranking model placed a specific flight first and translating that into plain English ('This flight saves you 2 hours and costs only €14 more'), which simultaneously increases traveler trust and conversion. Level 2 is chatbots deployed across the entire funnel — top of funnel for inspiration, mid-funnel for hotel queries, and bottom of funnel for insurance questions on car rentals. He notes chatbots are most valuable not for task completion but as a learning signal: they reveal what travelers are confused about, enabling product teams to fix core UX. Level 3 is Generative UI — dynamically assembling a page in real time from user intent (e.g., 'skiing in the Italian Alps' triggers a map of mountains, weather, chalets, and flights on a page that did not exist before the query). Level 4 is Agentic Delegation, exemplified by Skyscanner's ChatGPT plugin (released the Friday before ITB), which lets users invoke Skyscanner inside ChatGPT via '@Skyscanner', refine queries conversationally, and follow through to live partner booking.

Skyscanner's internal AI platform abstracts model selection from engineers, allowing a single engineer named Louise to build a World Cup Football Finder feature — enter your team (e.g., 'Brazil'), see all matches, then book flights and accommodation — in hours using Claude Code and spec-driven development. After a week of design polish, the feature attracted 140 million viewers, illustrating a new product velocity that Sierra calls 'not what Skyscanner is used to.' On the partnership side, Sierra warns that a future where every company runs its own agent polling airline prices every 2–3 hours is technically impossible given real-time dynamic pricing requirements. Skyscanner's response is an open API layer — REST and MCP endpoints — covering live flight, hotel, and car prices plus ML-derived intelligence (booking timing signals, route demand elasticity, user intent clustering). Current partners include Microsoft, Samsung, and Yahoo, plus a growing startup ecosystem.

Sergio Golia (Amadeus) challenges the 'plug in a chatbot and travel gets better' narrative directly, arguing that the real 2026 challenge is not building a clever AI but making AI reliable, trustworthy, and deeply integrated into dynamic travel ecosystems. He demonstrates two use cases in under three minutes each. The first is an airline marketing optimization tool: an airline logs into an Amadeus agentic platform, which has already processed load factors, campaign performance, and hotel occupancy data to surface three underperforming routes. Selecting Chicago, the system quantifies the opportunity, proposes campaign options with expected impact, channel attribution, and budget, generates an agency-ready creative brief, and can autonomously publish the campaign — delivering a 'double-digit uplift on advertising spend' in the demo. The second use case is a live, on-stage voice AI demonstration: Golia calls an AI assistant in real time (in a noisy ITB hall), changes a flight from Berlin to Nice from March 5 to March 8 (9:00 a.m. economy, fare difference €32.72), pays with a stored card ending in 1019, and receives an email confirmation — all in under three minutes without human intervention. He highlights that behind that single call runs a multi-agent architecture: one agent for authentication, one for flight change, one for payment processing, all coordinating invisibly. The AI also reads emotional context, knows when to escalate to a human agent, and automatically cascades rebooking to seat assignments, baggage, and connected transfers.

Anne Pruvot (SNCF Connect & Tech) provides the national rail operator perspective. SNCF Connect has 17 million customers, 26 million users, 1.6 billion site visits, and sold 233 million tickets in 2025, making it France's largest e-commerce platform and the first site where many French consumers used a credit card online. She presents three use cases at different maturity stages. First, a live production AI chatbot: originally built as an internal tool for customer advisors to retrieve information faster, it was extended to the consumer-facing chatbot. Results: 95% of chatbot interactions require no human follow-up, customer satisfaction on written channels improved by 20 points, and chatbot usefulness more than doubled versus the previous version. SNCF Connect has been named France's 'Client Service of the Year' three consecutive times. Second, a ChatGPT plugin (submitted for review at the time of the conference, potentially live during the session) to explore how LLM-driven interfaces reshape booking behavior and to challenge internal teams to think in conversational product terms. Third, an innovation accelerator partnership with Hugging Face, Station F (Paris incubator), and HEC Paris business school, selecting five startups to tackle unsolved problems — including Orion (AI accessibility for visually impaired rail travelers) and Libra (AI-optimized bike placement at stations). Pruvot's closing argument is that AI creates value only when deployed with purpose: 'not just fast but meaningful, not just powerful but useful, not just impressive but trusted.'

All three speakers close by agreeing that AI makes their own jobs more complex — more channels, more edge cases, more team challenges — while making the traveler experience simpler. Golia's formulation: 'For me, not for our travelers. For our travelers it's getting easier.'