Marriott and Google Discuss AI Production Challenges
At the Skift Data + AI Summit 2026 in New York, Colin Coleman, Marriott's SVP of enterprise data, analytics, and AI, and Joff Romoff, Google Cloud's global head of travel and hospitality, joined a session moderated by Skift's Seth Borko, Skift reported. Both argued that launching AI pilots is relatively easy while scaling to production is the hard part, and they pointed to people and workflow barriers rather than model capability as the main blockers. Romoff cited examples reported by Skift: a large European hotel company cut call response time by about 75% through conversational call-data automation, and Air France reduced development time about 35% using Google Cloud's Gemini to centralize fragmented data sources. Coleman, who runs data across roughly 9,900 properties and 30-plus brands, warned that automating tasks as they exist can lock in broken workflows, saying, "This isn't about automation."
What happened
At the Skift Data + AI Summit 2026 in New York, Colin Coleman, SVP of enterprise data, analytics, and AI at Marriott International, and Joff Romoff, global head of travel and hospitality at Google Cloud, took part in an Ask Me Anything moderated by Skift's Seth Borko, Skift reported. Both framed the central enterprise-AI challenge as moving pilots into production, arguing that people and workflow constraints, more than technology, block that jump.
Reported results
Skift recorded examples attributed to the session:
- •Romoff said a large European hotel company cut call response time by about 75% through conversational call-data automation.
- •Romoff said Air France reduced development time about 35% by using Google Cloud's Gemini to centralize fragmented data sources.
- •Romoff put the average travel prompt at about 15 words.
- •Coleman said, "This isn't about automation," warning that automating a task as it currently exists can lock in a broken workflow rather than re-engineer for the desired outcome.
Technical context
Moving GenAI pilots into production commonly stalls on nontechnical bottlenecks: messy, distributed data and workforce practices built around legacy task boundaries. Centralizing data and building repeatable pipelines are frequent prerequisites before prompt or model work yields measurable ROI. Coleman, whose remit spans roughly 9,900 properties and more than 30 brands built on a large first-party data estate, has emphasized this data-architecture layer in related Skift coverage.
Industry context
Skift's reporting notes travel-sector variation, with cruise and aviation already converting AI into revenue while lodging's consumer-facing layer is still arriving. Sectors that standardize partner data flows and expose operational metrics early tend to realize faster, measurable gains.
What to watch
- •Adoption of centralized data fabrics and repeatable pipelines as a precondition for scaling.
- •Production metrics such as call-response time and development-cycle reduction.
- •Investment in workflow redesign and skilling rather than model tuning alone.
Key Points
- 1Senior Marriott and Google Cloud leaders said scaling GenAI from pilot to production is blocked mainly by people and workflow constraints, not model performance.
- 2Reported production results include a roughly 75% cut in call response time at a European hotel group and a 35% reduction in development time at Air France using Google Cloud's Gemini to centralize data.
- 3Coleman's caution ("This isn't about automation") reflects a broader lesson: automating existing tasks can preserve broken workflows, so redesign often precedes durable ROI.
Scoring Rationale
A practitioner-focused discussion from senior Marriott and Google Cloud leaders with concrete production metrics and a clear lesson that workflow and data architecture, not model capability, gate enterprise AI ROI. Useful and credible trade coverage, but a conference panel recap rather than a primary product or research event, placing it in the mid band.
Sources
Public references used for this report.
Practice with real Health & Insurance data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Health & Insurance problems
