I worked with a GM once who kept a laminated card behind the front desk. It listed every system that talked to the PMS, what each one did, and... this is the part that mattered... who to call at 2 AM when any of them stopped talking. She updated it every quarter. Her regional VP thought it was quaint. Her night auditors thought it was the most important piece of paper in the building. They were right.

Oracle just made a move that's going to matter to every hotel running OPERA Cloud, which at this point is a significant chunk of the industry. They've shifted their integration platform (OHIP) from a polling model to a streaming model. In plain English: instead of your connected systems constantly asking the PMS "anything new? anything new? anything new?" every few seconds, the PMS now pushes updates out in real time the moment something happens. Reservation change, room status flip, guest check-in... the data flows immediately to every system that needs it. They're using GraphQL Subscriptions and WebSockets, which (for the non-technical folks) is essentially the difference between refreshing your email every 30 seconds and getting a push notification the instant a message arrives. It's faster, it's lighter on the system, and it means fewer of those maddening moments where a guest checks in but housekeeping's system doesn't know for another three minutes.

Here's what I like about this. It's a real architectural improvement, not a marketing rebrand of existing functionality. The old polling approach created lag, ate bandwidth, and generated unnecessary server load... especially at properties with dozens of integrations all pinging the PMS simultaneously. With 1,200-plus partners building on the OHIP platform and 650-plus live in the marketplace, that's a lot of simultaneous conversations. Streaming cleans that up. And for properties where your infrastructure is already strained (and let's be honest, if your building was wired before 2010, your infrastructure is strained), reducing that constant back-and-forth polling traffic is meaningful. The real-time piece also opens the door for things like instant mobile key delivery, live housekeeping dashboards that actually reflect what's happening right now, and revenue management systems that can react to booking patterns as they unfold rather than on a delay. That's genuine operational value.

But here's where I start asking the questions that don't show up in the product announcement. Oracle is actively sunsetting their legacy SOAP-based integrations in favor of OHIP's REST-based APIs. That's industry speak for: the old way your systems connected is going away, and every vendor you work with needs to rebuild their connection to the new standard. If you're running OPERA Cloud with eight or ten integrated vendors... your door locks, your payment gateway, your housekeeping system, your RMS, your guest messaging platform... every single one of those vendors needs to migrate to the streaming model or eventually get cut off. Some of your vendors are Oracle marketplace partners with dedicated engineering teams. They'll be fine. Some of your vendors are smaller companies running lean, and rebuilding an integration isn't a weekend project. The timeline between "Oracle announces new architecture" and "your door lock vendor actually supports it" can be months. Sometimes longer. And during that gap, you're running a patchwork of old connections and new connections and praying they all play nice together. I've seen this movie before. The technology gets better. The transition is where things get ugly.

The other thing nobody's talking about: Oracle's cloud revenue just hit $8.9 billion in Q3 (up 44% year-over-year), and their remaining performance obligations are at $553 billion. That's not a hospitality number... that's the whole Oracle machine. Hospitality is a vertical inside a company that is aggressively, almost maniacally, moving everything to cloud subscription revenue. Which means the pressure to migrate every property off legacy systems and onto cloud-based, subscription-priced products is not going to slow down. It's going to accelerate. If you're still running on-premise OPERA and thinking you've got time... you have less than you think. The integration ecosystem is being rebuilt around OPERA Cloud. The partners are building for streaming APIs. The old architecture isn't getting investment. Nobody at Oracle is going to call you and say "we're forcing you to migrate." They don't have to. They just have to make staying where you are progressively more painful until moving is the only rational choice. That's not a conspiracy. That's how platform companies work. I've watched it happen with three different enterprise vendors over the last 20 years. Same playbook every time.

So here's the pitch: AI runs in the background, optimizes your pricing, handles 80% of guest inquiries, cuts food waste by 50%, speeds up housekeeping by 20%... and nobody gets fired. The "invisible manager." That's the framing from a new wave of coverage positioning AI as the silent co-pilot every hotel operator has been waiting for. The global AI-in-hospitality market is supposedly headed from $16.3 billion to $70 billion by 2031. And 77% of hoteliers say they're planning to throw 5-50% of their IT budget at it.

Let me tell you what actually happens.

I consulted with a 180-key select-service property last fall that bought into one of these "invisible" AI platforms. Conversational guest messaging, dynamic pricing recommendations, automated housekeeping task assignment. The demo was gorgeous. Worked perfectly on the sales rep's laptop. They signed at $1,400 a month. What the vendor didn't mention: the PMS integration took 11 weeks instead of three, required a middleware patch that nobody on the hotel's team understood, and the dynamic pricing module kept pushing rates that conflicted with the revenue manager's comp set strategy. The front desk staff stopped trusting the guest messaging bot after it told a guest the pool closed at 9 PM (it closes at 10) and offered a "complimentary spa upgrade" at a property that doesn't have a spa. The GM told me he spends more time babysitting the AI than it saves him. His words: "I didn't buy an invisible manager. I bought an invisible toddler."

Look, I'm not anti-AI. I'm an engineer. I've built rate-push systems. I understand what good automation architecture looks like, and some of what's emerging is genuinely impressive. The food waste tracking using computer vision in kitchen operations? That's real. The math works... if you're a 400-key full-service property with a serious F&B operation, you can see ROI in under a year. Voice-powered LLM systems that can handle multi-step guest requests? Getting better fast. But here's the thing nobody's asking: what percentage of the hotels being sold this technology actually have the infrastructure, the bandwidth, the staff training capacity, and the PMS architecture to make it work? The BCG-NYU report from last week quietly mentions that only 2.9% of hospitality workers have AI-relevant skills. The average hotel PMS is 15 years old. And 65% of North American hotels can't fully staff their existing shifts. So we're layering autonomous systems onto properties where the WiFi drops on the second floor and the night auditor learned the PMS from a three-ring binder in 2011. That's not an AI readiness problem. That's a fantasy-meets-reality problem. And I've been on the wrong side of that equation before... my first startup crashed because I built technology that worked perfectly in a demo environment and failed spectacularly in a real hotel at midnight. The gap between "works in the pitch" and "works at 2 AM when nobody's here" is where most of these AI promises will die.

The real question for operators isn't whether AI is useful (it can be) or whether it's coming (it is). The question is: does this specific product, at this specific price point, solve a problem my team actually has, on infrastructure my building actually supports, with a failure mode my least technical employee can actually recover from? That's the test. And Marriott's own SEC filing from early 2025 flags something even bigger... AI-driven platforms may shift bookings away from direct channels and loyalty programs toward intermediaries, potentially increasing distribution costs. So while vendors are selling you AI as a cost-saver, the macro effect of AI on the distribution landscape might actually cost you more on the top line. Nobody's putting THAT in the demo.

If you're a GM or owner being pitched an AI platform right now, do three things before you sign anything. First, ask the vendor what happens during a system outage at 2 AM with one person on shift. If the answer involves "contact support," walk away. Second, get the actual total cost... not the monthly subscription, but implementation, training, integration maintenance, and the productivity dip during the transition. That "$500 a month" system has a very different real cost. Third, demand performance data from properties that match yours... not the 500-key resort with a dedicated IT team, but the 120-key select-service with a night auditor who's also watching the door. If they can't show you that, they haven't proven their product works where you need it to work.