A next level of Customer Centricity : AI-enabled Synthetic Personas in Hospitality Innovation

This session at ITB Berlin 2026 explored the next generation of customer understanding in hospitality: AI-enabled synthetic personas and digital twins. The session opened with a keynote by Sven Dörrenbächer (former CMO/CDO, now AI strategist) followed by a panel moderated by Kirsten Feld-Türkis, featuring Julie White (CCO Europe & North Africa, Accor Hotels), Ali Beklen (founder/managing partner, Hotel Runner), and Dr. Joerg Meurer (partner/MD, Profit Germany strategy consultancy).

Sven framed AI as the sixth transformational wave in history — after the printing press (1440), steam engine, electricity, automobile, and internet — arguing each divided the world into 'those who moved and those who were moved.' He stressed holding two seemingly contradictory stances simultaneously: embracing opportunity while remaining paranoid about risk. His central point: 'AI is not wise. AI often lies. Your reputation is on the line.'

On productivity, Sven cited Klarna's AI assistant (launched 2023, results shared 2024): it handled 2/3 of all customer service chats, covering 2.3 million conversations across 23 markets in 35 languages, equivalent to 700 FTE, with an estimated first-year profit impact of $40 million USD. However, he noted a cautionary sequel: a year later, Klarna saw service quality drop and customer complaints rise, and began rehiring humans — illustrating that 'AI doesn't replace empathy.' He also cited McKinsey data that 80% of companies are still not achieving productivity gains with AI, attributing this to leadership failure: AI success is only 10% a technology question and 70% a people, culture, and process question.

Dr. Joerg Meurer defined synthetic personas as LLMs configured to take on the personality of a target guest segment — fed with demographic, psychographic, lifestyle, and travel-need data — creating a '24/7 crystal ball' that answers questions on demand. He drew a distinction between synthetic personas (relatively static, fed once and updated periodically) and digital twins (dynamic, continuously learning from real-time data feeds). A key case study from Profit: they created a full-fledged brand playbook for three different hospitality brands in just two weeks — work that would normally take two months. Another case study: a client had three guest personas that all showed affinity for 'beach and garden games' (a generic survey finding). When the synthetic personas were prompted to generate specific game concepts suited to each segment, they produced ideas like 'glow badminton' and 'a specific ponk ritual' — creative, differentiated outputs that traditional research processes would not have generated. Dr. Meurer also recounted a client who initially learned about synthetic personas through consulting, but then built their own persona independently; when the client described their own hands-on interaction with the persona, their strategic thinking had measurably transformed — demonstrating that doing beats learning about.

Julie White described Accor's persona work across its 9,000+ hotel portfolio spanning economy (Ibis) through premium and boutique (Handwritten, Adagio, Tribe). Personas are built around existing guests and target audiences, used for targeted marketing, brand experience differentiation, and CRM personalization, with separate loyalty personas tracking behavior across brands. In AI-enabled applications, Accor uses synthetic personas to test e-commerce concepts — new landing pages, CRM targeting, personalization flows — enabling simultaneous testing across multiple markets, languages, and persona combinations at speed and scale. Crucially, White emphasized that Accor always finalizes testing with real guests and loyal members before deployment: 'We use them to scale. We use them to go to speed. But so much is important about the trust in our brands.' She also announced a major milestone: Accor was the first direct-channel hospitality brand house to integrate with ChatGPT via their Accor app, developed through a proactive collaboration with OpenAI. Bookings through the LLM channel are currently in 'single digits but growing exponentially.' A key content insight from this integration: in LLM-mediated search, adjectives matter far more — 'not just does it have a pool but does it have a heated pool,' 'not just a meeting room but a meeting room with natural light.'

Ali Beklen (Hotel Runner) addressed the data infrastructure dimension. He noted that C-level executives in Fortune 500 companies now make 300 decisions per day on average, versus 15–20 decisions per day 20 years ago — a 20x increase. Average consumers make 3–5 decisions per day currently, but AI will bring them to a comparable acceleration. He argued that if hotel organizations are not ready to handle 30 consumer decisions per day at 'light speed,' they will lose demand. Hotel Runner used synthetic personas to address hospitality's post-pandemic talent shortage by creating digital twins of hotel staff roles — revenue, sales, housekeeping — creating a 'trainingless environment' that allowed hotels to onboard and scale without the traditional training bottleneck. Beklen introduced the concept of LLM-to-LLM communication: if guests gave permission for their personal AI agent to share preferences with a hotel's internal LLM before arrival, the hotel system could proactively notify every staff member of the individual guest's specific expectations — making segmentation itself potentially obsolete in the near future.

The panel concluded with practical recommendations: start by building a custom GPT with your own customer data immediately; prioritize data quality and content optimization for LLM-readable formats; continue testing with real guests alongside synthetic ones; and adopt a 'reset' mindset — reset your timelines from two-year horizons to three-to-six-month delivery cycles.