Is Traveltech ready for AI? How to feed the Arficifial Agents
Summary
This panel session at ITB Berlin 2026, moderated by Dirk Rogl (Managing Director, Travel Commerce GmbH), examines whether the travel technology stack is ready for AI agents. The two main speakers are Mathis Boldt, CEO of GIATA (formerly spent 20 years building Booking.com and GetYourGuide), and Bradley Johnson, VP of Product Management at Sabre.
Mathis Boldt opens with a provocative framing: 'AI is only as good as the underlying data — and in travel, data equals expectations.' He argues that the industry has been promising a 'connected trip' and 'personalization' since 2012 without delivering on either, and that the core problem is a foundational data challenge, not an AI capability gap. He illustrates this with a vivid case study: a wealthy Canadian family booked a wedding venue and hotel 15 kilometers apart, only to discover on the day before the wedding that a mountain made the actual road distance 80 km — because a single data point was mathematically accurate but operationally catastrophic. He uses this to argue that entrusting a €10,000 travel booking to an AI agent requires far higher data accuracy than ordering coffee beans online.
Boldt describes GIATA's product suite addressing these gaps. GIATA Multicodes assigns every hotel in the world a universal 'GIATA ID' (analogous to a social security number or ISIN) to eliminate duplicate and mismatched property listings across supplier databases. Room type mapping solves the problem of a single hotel having 130+ different room type combinations across 9 suppliers, making AI-driven rate comparison impossible without normalization. GIATA Drive is a 'source of truth' platform where hotels can self-manage granular property data (facts, amenities, room types, images) and distribute it in formats including JSON compatible with MCP, making hotel websites directly machine-readable for agentic AI. GIATA also acquired a company called Smart Seere, an AI decisioning platform that profiles website visitors in milliseconds to predict intent, willingness to spend, and churn likelihood, then matches them with the right offer at the right funnel stage (e.g., showing vouchers to early-stage visitors who are about to leave, sponsored listings to mid-funnel undecided users, and clean product pages to ready-to-book visitors).
Bradley Johnson presents Sabre's perspective on the full agentic commerce funnel: discovery, retailing, checkout, and servicing. On discovery, Sabre is building a normalized product catalog mixing static and fresh-cached data to deliver low-latency, low-cost results, noting that existing retail commerce protocols lack the structured data and relevancy travel requires. On checkout, he emphasizes keeping the customer in-channel all the way through secure payment, fraud detection, and supplier reconciliation, using 'deterministic workflows' that sequence required steps with configurability for partner needs. On servicing, he identifies it as 'the biggest opportunity when it comes to Agentic AI in terms of driving out cost,' specifically citing proactive disruption sensing — e.g., an 80% flight disruption probability alert — as transformative for traveler experience.
Sabre's flagship live case study is a newly announced partnership with Mindrip and PayPal, demonstrated live on stage: a conversational travel booking flow where a user asks 'What are the best dates to fly from New York to Alaska?' receives the answer 'Your ideal window is late August to mid-September,' then books through Sabre's shopping APIs, Mindrip's agentic UX, and PayPal for payment. Johnson confirmed this is a real product, not a demo concept, though full availability is described as 'very soon.'
In the moderated Q&A wrap-up, Boldt cites a recent GIFT study finding only 2% of travelers today are comfortable letting an AI agent make purchases on their behalf, and stresses that the industry must continue serving customers across all channels — offline agents, tour operators (which he notes are still growing despite decades of predictions otherwise), and AI. Both speakers converge on 'human in the loop' as the key trust mechanism for moving agentic commerce down the funnel from inspiration to transaction. Both predict agentic travel bookings are 'months away' rather than years, with Boldt adding that higher-value summer vacations will lag behind lower-stakes corporate travel.
Key Takeaways
- •Data accuracy is existential for agentic travel: Mathis Boldt argued that in travel, 'data equals expectations,' and when data breaks, 'experience breaks, trips break.' He illustrated this with a real case where a wedding party discovered the night before their ceremony that a hotel listed as 15 km from the venue was actually an 80 km drive due to a mountain — a data point that was mathematically correct but operationally catastrophic for a €10,000+ booking.
- •Only 2% of travelers are currently comfortable letting AI agents make purchases on their behalf: Boldt cited this GIFT study statistic to argue that the industry is still in the early agentic adoption curve and must continue serving customers across all channels rather than assuming a wholesale shift to autonomous AI purchasing.
- •GIATA Multicodes solves hotel identity fragmentation by giving every hotel in the world a universal GIATA ID, similar to how ISINs or EAN numbers work, so AI agents can unambiguously identify and compare properties across all supplier databases — a foundational requirement for any meaningful agentic price or availability comparison.
- •Room type name chaos makes AI comparison impossible without normalization: A single hotel with five or six room types can have 130+ different room type name combinations across nine suppliers due to legacy and marketing reasons. GIATA's room type mapping collapses these into a canonical set, enabling AI agents and businesses to find best rates on equivalent room types across all suppliers.
- •Hotels have lost the narrative of their own properties: Boldt warned that without hotels owning and publishing structured data, AI will resort to image recognition and scraping to infer amenities — producing fuzzy probabilistic answers that are acceptable for guessing pool sizes but dangerous for location data. He gave an example of a supplier's metadata mixing address details from a Barcelona hotel with an address from a South American hotel in the same record, which would send half the bookings to the wrong country.
- •GIATA Drive distributes hotel property data in JSON with MCP compatibility, enabling hotels to make their own websites directly readable by agentic AI without requiring screenshot parsing or third-party inference. Boldt positioned this as how hotels reclaim narrative ownership in the age of LLMs.
Transcript
And ladies and gentlemen, we coming to the session I'm really looking for uh just because it's the core of our track today. Are we ready for AI? And just as a very short wrap-up for from from our yesterday's AI track. Anka is saying referring to uh to to to Airbnb, it may take three years since we see aic shopping. When we listen to to our dedicated speakers from yesterday, it takes a few months just because everything is available. Instant pay instant payment is available. Agentic shopping infr...
