PwC's AI Agent Survey of 1,000 US business leaders found that 79% of organizations say they have adopted AI agents to some extent. Deloitte's 2026 State of AI in the Enterprise reports that only 11% have agents running in production. Same year, similar populations, almost seventy points of daylight between adoption and production.
Hospitality has its own version of the same gap. The 2026 Hotel PMS Impact Study found that 98% of hotels have begun using AI in some capacity, but only 32% have it embedded across most of their operations. Two-thirds of properties are running pilots that haven't made it to the operational loop.
If you run a property and your AI initiatives feel like a series of half-finished experiments, you are not behind. You are the median. The question is why pilots stall, and what gets one to production.
What "in production" actually means for a hotel
In production means an AI capability is running in the live operational loop without a human babysitting every interaction. The front-desk agent handles inbound calls in off-hours. The revenue manager's pricing engine moves rates without manual sign-off. The guest messaging bot books restaurants, not just suggests them. A pilot is none of those, it is a demo that has not been integrated.
The difference is not effort, it is integration. A pilot lives in a sandbox alongside the operational systems. A production agent lives inside them, reading from the PMS, writing to the CRM, transacting through the booking engine, and authenticating against the property's data layer. Most pilots stall because the second piece never gets built.
The single biggest reason pilots stall
The Anthropic State of AI Agents 2026 report names integration with existing systems as the primary challenge for 46% of respondents. Phocuswright's hospitality-specific Budgets, Barriers, and the Race to Agentic AI survey reaches the same conclusion for travel: integration complexity, talent gaps, and data security top the barriers. Not lack of interest, not lack of funding.
Hotels have an especially hard version of the integration problem. The average independent property runs a PMS that was built before AI agents existed. It exposes a partial API or none at all. The CRM is a different vendor. The channel manager is a third. The booking engine is a fourth. Getting an AI agent into production means stitching the agent to all four, and the integration work is unglamorous, expensive, and often blocked on the PMS vendor's roadmap.
Properties on modern cloud PMS infrastructure have a measurably easier time. Oracle's Hospitality Integration Platform alone exposes 3,000+ APIs with 1,200+ integration partners. Properties on legacy on-prem systems have to do the work twice, once to get data out, once to wire results back in.
The pilots that do make it to production
The pilots that ship share three characteristics. The use case is bounded, one job, not the whole guest journey. The integration surface is one system, not four. The success metric is operational and measurable in weeks, not strategic and measurable in years. The 35-room boutique that ships a voice agent in four weeks does all three.
A 35-room boutique in the HotelTechReport sample deployed a voice AI agent connected to its PMS in four weeks. It now handles 60%+ of inbound front desk calls. The pilot was scoped to one job (inbound calls), one integration (the PMS), and one measurable metric (call deflection rate). It made it to production because it didn't try to do everything.
The pilots that stall are usually scoped the other direction. "An AI concierge that personalizes the entire stay." That is six integration surfaces, eight stakeholders, a fuzzy success metric, and a 12-month roadmap. It does not survive its first quarter unless the property has internal engineering muscle most independents do not have.
What the data says about the next twelve months
Phocuswright's Budgets, Barriers research found that 61% of travel businesses are experimenting with or scaling agentic AI, but only 6% are actively scaling, and 22% are beginning to scale. That leaves 35% stuck in experimentation, which lines up almost exactly with the 32% of hotels that have AI embedded across operations per the HotelTechReport study.
Generative AI is the #1 tech investment priority for the next twelve to eighteen months across travel businesses surveyed. The budget is there. The pilots are running. The bottleneck is integration. Which means the operators who solve the integration problem in the next twelve months are the ones who graduate from the 35% experimenting cohort into the 6% scaling cohort.
What independents should actually do
The 6%-scaling cohort is dominated by chains. Independents catch up by scoping pilots tightly, not by waiting for the perfect platform. One job, one integration, one operational metric. Ship that, learn, pick the next bounded pilot. The properties that ladder through three of those in a year end up in the production cohort. The "AI everywhere" initiative does not survive its first quarter.
The starting move is the system the property already controls (the PMS, usually) and the use case it most directly enables (front-desk deflection, after-hours guest messaging, basic upsell prompts). We covered the discoverability piece in Boutique Hotels and the Coming AI Discoverability Layer and the booking-flow piece in The Agentic Booking Shift: What Hoteliers Need to Know Before 2027. The pilot-to-production work sits between those two.
Gartner forecasts that 40% of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5% in 2025. The wave is happening on the calendar, not on the operator's timeline. Properties that are running bounded pilots now end up in the production cohort. Properties waiting for the platform play do not.
If you want to walk through which pilots are actually shippable for your property in the next quarter, reach out. We are working with a handful of operators on exactly this question right now, and the answer is almost always "smaller than you think, sooner than you think."