Is ARI plumbing the main impediment to Agentic AI?

Yes, the existing piping is not ideal. Remember, a bunch of hotels in Europe still use Booking.com as their booking engine, because there's no piping at all (no booking engine of their own, or their PMS is premise-based and lacks APIs).

Google was able to pull it off (their Hotels product represents 1.2 M hotels). Others can do it. The question is: do they want to?

Yes, outdated ARI plumbing is a huge immediate obstacle, but it isn't the only one.

Legacy hotel systems rely on slow, cached data and one-way connections that frequently crash when AI agents try to make real-time bookings. However, even if the technology is fixed, bigger business problems remain: hotels fear losing customer relationships and revenue margins to AI intermediaries, and payment liability for AI errors remains unclear.

In short, while the technical pipes are broken, the deeper issue is a lack of trust, clear business models, and secure payment systems for autonomous bookings.

The main impediment isn’t tech plumbing; it is inherent in the concept of “agentic AI” itself. It is based on Gen AI LLMs, which are probabilistic, next-best-word engines. They have no “real-world” models, eg understanding higher / lower, before / after, logic, distance, numbers.  They mimic very well.  Super for wordsmithing and unstructured data / document analysis, poor for most everything else. Add a sequence of AI agents and any small mistakes get magnified.

The newer LLMs have a huge, symbolic, deterministic (if-then-else type) harness wrapped around them to fill gaps, like deciphering Excel formulas, and running Python code to count the e’s in “Our TRevPAR strategy is hoping everyone orders the £24 G&T.”

There are few processes in hospitality that can tolerate these mostly right, occasionally very wrong decision engines. A hotel can’t check in the right customer into the right room most of the time, or charge the right customer the right folio amount most of the time, or accidentally commit to a group booking for 100 people at £15 per room, rather than for 10 people at £150 per room.

Leave “agentic AI” based on LLMs for internal use, like filtering guest feedback, but have a human vet the output.

No. The plumping is a serious issue but I would argue that the bigger impediment to Agentic AI lays more up-funnel in the AI citation or search visibility landscape. Because if an AI can't crawl your website or if it can't find consistent information or if the information isn't crystal clear, the result is fewer recommendations for the hotel and more commerce reverting to larger, more seemingly trustworthy third parties like the OTAs.

The plumbing of ARI is downwind of that. Still vital, but we have to think about what's going to first help get more eyeballs and robots on the direct channel in order to achieve the key commercial goals of first filling the hotel and then filling it with direct bookings to optimize for net revenues. With those two solved, then that creates some buffer in the P&L to reinvest in the solutions, vendors, resources and just time to think through how to solve the ARI connector challenge.

Personally, I would argue the plumbing is a real constraint, but not the main one.

Thin pipes are a symptom. Cached ARI, one-way APIs and layered PMS, CRS and channel manager stacks exist because hotels built distribution around OTAs and GDSs, not conversational, real-time demand. Widening the pipes will help, yet an agent connected to inconsistent data simply makes mistakes faster.

The deeper impediments sit elsewhere:

- Data integrity: rate codes, restrictions, room types and policies are rarely consistent across systems. Agents need structured, trustworthy content, not just availability.

- Commercial governance: parity obligations, channel costs and ownership of the guest relationship remain unresolved. Few hotels have decided what they want agents to sell, and on what terms.

- Trust and accountability: payment, modification, cancellation and liability when an agent books incorrectly are still largely undefined.

- Operational readiness: the property must deliver what the agent promised, at arrival.

MCP, A2A and UCP will mature, and connectors will multiply. The hotels that benefit will treat agentic AI as a distribution and operating model decision, with technology following strategy.

The pipes matter. What flows through them, and who governs it, matters more.

First of all, let’s make a clear distinction between the agentic AI platforms recommending a hotel and enabling a hotel booking.

The AI platforms are like gigantic vacuum cleaners, sucking content about hotels from all available public sources: directories, main data providers, OTAs, review sites, and yes, the property own website if optimized for AEO.

Enabling a hotel booking is much more difficult and to do this the AI platforms have to connect to whoever owns the last ARI mile: CRS, PMS, Channel Managers, GDS and OTAs, as well as payment solutions. 

Why did Meta partner with Expedia for its new Muse Agentic AI platform? With one B2B connector, Muse gained access to 750,000 hotels, 400 airlines, 150 car rental companies and 2.9 million short-term rentals.

Google unveiled its agentic hotel booking functionality in AI Mode with OTAs like Booking..com, Expedia, Trip..com and Major Hotel Chains like Choice Hotels, Hilton, IHG, Marriott and Wyndham.

At this point, the Agentic AI platforms are mere entry doors to the OTAs and major hotel chains. 

Independent hotels? The only option is for their CRS, Channel Managers, PMS vendors to plug into the agentic AI platforms via Google’s UCP or MCP or A2A, allowing travelers to book rooms.

It's only early days and there is much technical capability to come with the next generation of connectors.

Summed up, digital availability, accuracy and speed of delivery has been the grail for our industry for a very long time. That is not going to disappear any time soon.

This information is retained in operational systems for logical reasons and will remain the case in the future. For rooms and every other product.

If reliable information about product availability cannot be surfaced efficiently the same downfall this has caused in the pre-AI age remains.

Ultimately, it is an interface. It is only as good as the information it can source and transpose.

The integration gap is real, but only part of the answer. An independent hotel cannot expose live rates to an agent until its booking engine or channel manager vendor ships the integration, the chokepoint that stalled Book on Google. But an agent books only a hotel it has found and chosen, so this sits at the end of the funnel.

Hotels get chosen in the answer. AI Overviews tend to cite the page that settles a near-booking question, like the pet fee or late check-in cut-off. When your site, booking engine and OTA listing disagree, the structured OTA listing is often the one cited, and no connector changes that. The fix is dated policy pages with structured data, consistent across channels.

Then there is measurement. Assistant sessions without a referrer land in GA4's Direct channel, and Search Console's generative AI report shows Google impressions with no query or booking attached. Without seeing agent-driven discovery, you cannot tell whether a connector produces bookings.

Hotels running metasearch already send Google Hotel Center the ARI feed that live prices in an assistant depend on. For them the plumbing is further along than the question assumes; their first obstacle is being citable and measurable.

Connecting one's ARI to AI channels should be done. OTAs are already doing it for you and of course because they have massive and standardized inventory and data, they're preferred options for the AI bots for now. Hotels should plug in ASAP that's the minimum.

The real issue is how to supply lots and lots of structures data using Schema markup and other structured methods to really give AI bots a lot of context about your hotel. Not just "Spa: Yes" but all the data about your Spa, the spa menu, the number of treatment rooms, the prices, the brands you work with, reviews, and more.

This is where hotels have a chance to stand out, because OTAs wont have as much context and AI likes more context.

No; plumbing is a constraint, but it's a solvable one. ARI is, in a way, a commodity, already available to OTAs, metasearch and GDS at scale through the same PMS, CRS and channel manager stack.

What agents change is the pattern of use: First, look-to-book: agents will query far more than they book, and cached ARI and one-way APIs weren't built for that load. Second, write access: holding, modifying and servicing bookings is harder than reading availability. Third, rate integrity: agents compare prices in real time, so any parity gap shows instantly. These are engineering problems that some of the new protocols (UCP, MCP, A2A) will largely resolve.

The new elements are:

Content, not only ARI: Agents rank on structured content, reviews, brand signals, and then price. A connected hotel with weak content stays invisible.

Guest ownership. Is the platform going to send users to the direct channel, or the OTA? Or try to process directly?

Pricing intelligence. Hotels need to see what agents see and stay competitive in that channel.

Demand. Agentic bookings are still a small share of volume, but growing extremely quickly.

The winners will have clean ARI, market-aware content and pricing, and direct as the best-value path for agents too.

No. ARI connectivity is merely a symptom of a much bigger structural problem.

Hotel technology was designed to distribute rooms and rates, not to enable autonomous machines to execute complete commercial transactions.

Even with perfect API connectivity, AI agents need real-time availability, consistent offers, cancellation policies, secure payments, guest identification, loyalty recognition, and reliable booking confirmations.

But there is another, potentially bigger challenge: commercial aggregation.

Imagine asking ChatGPT:

“Find the best family hotel in Munich under €300 per night.”

Do we really expect an AI agent to connect individually to 800 different hotel MCP servers, retrieve availability, compare rates and policies, and complete a booking?

Of course not.

AI agents need standardized, comparable inventory accessible through aggregated platforms. This is precisely where OTAs already have a massive competitive advantage.

Technologies like MCP, A2A, and UCP improve connectivity, but connectivity alone doesn't create a competitive distribution ecosystem.

My biggest concern is that hotels confuse connectivity with competitiveness.

Without an open, standardized direct-booking infrastructure, Agentic AI will not eliminate intermediaries. It will simply create new ones.

The strategic question isn't who connects to AI first. It's who controls the distribution, the guest data, and ultimately the customer relationship.

For most hotels today, yes. But I would put it differently: the pipes aren't too narrow, they were never designed to be open in the first place.

Many PMS, CRS and channel manager systems were built to keep data inside. Cached ARI and one-way APIs reflect that. B2B AI connectors can bridge some gaps, but cannot remove the limitations of the underlying systems.

AI agents need to do more than retrieve availability or recommend rates. They need to modify reservations, handle payments and coordinate tasks across systems. That requires real-time, bidirectional access to data and the ability to act, not just look.

Open, API-first platforms are built the other way round. They provide the foundation for agents to access live hotel data and take action through MCP servers. Standards like A2A take this further by enabling agents to work together.

Autonomous hotel operations won't come from adding more connectors. They require open infrastructure with secure governance, giving hotels control over their data, which agents they use and what those agents can do.

The real challenge isn't fixing the pipes. It's opening up the entire infrastructure.

No. The pipes are thin, but they are not the blockage.

ARI pipes were sized for a different traffic pattern: channels pulling cached rates through one-way feeds, at look-to-book ratios someone had already agreed to pay for. Agents break that. They query more often, in more combinations, and expect live answers. Caching and aggregation can absorb that load, for a price.

Pipes get rebuilt when the traffic through them pays. The technology is not the blocker. Three unsettled questions are. Who absorbs the query cost: hotel, vendor, or aggregator? Who owns the transaction: payment, cancellation, refund, liability? And who owns the guest when the agent holds the relationship?

In my earlier PMS 2030 viewpoint, I anticipated that PMS, CRS, and channel manager would collapse into one real-time core. Agentic demand pays for it. Infrastructure readiness decides who wins. Incentives decide who gets ready.

Fix the incentives and the pipes will follow.

ARI plumbing is a prerequisite, but I think calling it the main impediment frames the opportunity too narrowly.

Making a room available to an AI agent does not automatically make the hotel better at selling the stay. Rates and inventory tell an agent what is available. They say much less about why a guest should choose that hotel, which experience fits their trip, or what else would make the stay more valuable.

A couple planning an anniversary is buying something different from a business traveler arriving late. Exposing the same room inventory to both does little to address that distinction. The opportunity is to connect guest intent, relevant experiences, and a seamless purchase, with the hotel retaining the context and relationship.

That work can start on the hotel’s own website today. Understanding what brought someone there and carrying that intent through booking creates value now and a stronger foundation for agentic commerce.

Yes, fix the pipes. But hotels should ask what flows back besides a reservation. If an AI platform understands the guest while the hotel receives only dates and a room type, we risk recreating the same distribution dependency through a new interface.

It’s not only plumbing. It’s business logic designed for humans.

The easy answer is to scale the pipes, but that is not enough. Every system in the chain, website, CRS, channel manager and PMS, would need to scale at the same time, and someone has to pay for millions of searches that never become a booking. Blocking agents is not an option: blocking agents means blocking future guests.

For me, the main problem is the business logic. Our processes were designed for a person who searches, decides and books once. Agents will not only use our systems more, they will use them differently. An agent can book, cancel and book again every time the price goes down, like trading on the stock market. Our cancellation policies, overbooking and revenue management were not designed for that.

Agent-to-agent is not the solution by itself: if my agent works on the same pipes and rules, the problem is still there.

We need to rethink the business logic for agents as customers, with things like holds that expire or cancellation rules made for machines. The question is not how big the pipes are, but which rules we apply when the customer is a machine.

I wouldn't put ARI plumbing at the top of the list. A hotel first has to give the traveler's agent a reason to consider it.

That starts with accurate, detailed property information, with the hotel establishing itself as the authoritative source. It also means paying attention to the experience commons: reviews, social media, and traveler photos that help agents assess whether the property actually delivers what it describes. Hotels can improve the experience, but they don't control that independent evidence. If an agent can't establish fit and trust, the hotel may never get the chance to present an offer.

Then there's the offer itself. ARI plumbing that just reproduces the room type/rate plan combinations available through an OTA or GDS is table stakes. Making the same offer accessible through another channel gives consumers little reason to change their behavior.

The opportunity is to use the richer intent from the traveler's personal agent to create a more attractive offer: the right room, services, benefits, attributes, and terms for that particular trip. That's what can pull consumers toward direct agentic bookings. And, of course, the hotel has to deliver what it promises.