How AI is Transforming Business Travel Across the Journey
Summary
This panel at ITB Berlin 2026 brought together four business travel AI practitioners — moderated by Claudia Unger (VDR) — to examine how AI is reshaping corporate travel across booking, ground transportation, meetings/events, and data infrastructure. The session opened with a telling audience survey: fewer than half of attendees had piloted even one AI feature in their travel program, which the moderator used as a benchmark for the room.
Olivier Garbé from MXGBT (now AGBT) described a three-layer AI architecture: a technology platform layer, a data fuel layer, and a split between machine learning for recommendations and LLM-driven decision-making for automated workflow execution. His team has deployed AI chat agents for clients including Aensia for over a year. Two hard metrics stood out: an AI chat agent reduced human contact rate by 23%, and automated policy rate recommendations generate an average saving of $61 per booking.
Sebastian Adams from Uber for Business highlighted Uber's scale — 33 million trips and deliveries per day — as the raw material that feeds AI-powered route optimization, reliability improvements, and cost efficiency. He emphasized that autonomous vehicles are no longer hypothetical: live AV pilots are already active in Texas via the Uber app (partners include Waymo and others), and Adams boldly predicted AVs will arrive in Germany within the next 12–18 months. He also noted that by end of 2026, one in four employees will be Gen Z, accelerating demand for intuitive, on-demand, app-first services in the B2B world.
Alexander Schott from Mitago (35 years in travel IT, previously Cytric/Amadeus) focused on the MICE RFP cycle. He described 'negotiation memory' — an AI capability that learns a corporation's historical negotiation patterns and surfaces only the venue offers most likely to close, eliminating noise. He cited a partnership with Hiva AI, which automates hotel-side RFP responses: hotels using Hiva respond 7.5 hours faster than non-users and close at double the rate. Schott drew a clear line between what is realistic today (intent-driven search that can execute, not just recommend) and what remains hype (fully unattended AI systems without human oversight).
A significant portion of the panel focused on what Garbé called 'unsexy work': data quality, model drift monitoring, and guard rails. AGBT explicitly avoids hyperpersonalization and does not train models on personal data due to banking regulatory constraints. They address hallucinations by forcing the model to say 'I don't know' rather than fabricate, and use both offline and online evaluation loops with automatic or manual retraining.
On interoperability, Schott argued the travel industry does not need more content APIs — it needs transactional APIs that let AI agents execute tasks, not just retrieve information. Garbé advocated for MCP and A2A (agent-to-agent) protocols, calling them effectively 'API version 3 and 4,' and argued NDC is 'everything but a standard.' The panel acknowledged the current fragmented API landscape as a core blocker for agentic AI in travel.
The Q&A covered LLM selection strategy (Garbé: evaluate on pertinence, speed, and cost — sometimes SLMs are appropriate; Schott: start within your governance framework and just begin), the AV protocol landscape (MCP vs. A2A vs. Google's UCP), and practical advice for hotels paralyzed by decision fatigue. Schott coined 'tool zoo' — the failure mode of starting many disconnected AI initiatives that don't interoperate.
Key Takeaways
- •AGBT's AI chat agent cut the human contact rate by 23% for at least one major client (Aensia, deployed over a year ago), and automated policy rate recommendations deliver an average saving of $61 per booking — two concrete metrics Garbé shared as proof of commercial-grade AI ROI.
- •Uber processes 33 million trips and deliveries per day, and uses that data volume as the primary input for AI-driven route optimization, reliability, and cost efficiency — framing scale itself as the competitive AI moat in ground transportation.
- •Live autonomous vehicle rides are already bookable on the Uber app in Texas today (via Waymo and other partners). Adams predicted that Germany will see commercial AV deployments within 12–18 months, comparing the shift to the transition from horse to car.
- •Hotels using Hiva AI's automated RFP response tool respond 7.5 hours faster than non-users and close at double the rate — Schott cited this as the clearest business case for AI adoption in the hotel/MICE segment.
- •Schott drew a concrete line on what AI can do today vs. hype: intent-driven search that executes ('25 high-executives, confidential, within 2 hours of HQ — close it') is live. Fully unattended AI systems without human oversight are not yet ready.
- •Schott coined 'negotiation memory' — an AI system that ingests a corporation's historical RFP negotiation patterns and uses them to filter venue shortlists to only those likely to close, reducing late-stage RFP failures.
- •AGBT (a bank-regulated entity) explicitly does not train AI models on personal data and avoids hyperpersonalization. Hallucinations are addressed by forcing the model to respond 'I don't know' rather than invent an answer, plus continuous offline/online evaluation with automatic or manual retraining for model drift.
- •Schott argued the travel industry's biggest API gap is not content but execution: 'We need transactional APIs because if you want to send out a fleet of bots or agents that do things on your behalf, you have to let them execute.' He called on all software vendors to slice systems into components exposed via APIs.
Transcript
Uh our second session is how AI is transforming business travel across the journey. So if you're thinking of the trip the traveler when they are booking when they are traveling when they come back all the things that they have to go through all those are touch points potentially for an AI um inclusion and so that's what I want to discuss with some panelists today from uh various areas of business travel and AI really is not just a buzzword it's it's shaping our industry and what we have found wi...
