SHEYU.AI舍予基业
AI Daily Brief

Microsoft's Oracle AI cloud exit is a lesson for hotel AI customer service budgets

Microsoft's decision not to renew an Oracle cloud contract shows why hotels should price AI customer service by the front-desk work it absorbs, not by model access.

·4 min read
Microsoft's Oracle AI cloud exit is a lesson for hotel AI customer service budgets
AI-generated illustration, not a news photograph

Microsoft has decided not to renew a cloud computing contract with Oracle, while tightening oversight of other data centre projects and exiting some it deemed higher-risk, according to a report by Phoenix News. The same report said some key figures in Microsoft management were unhappy with the high cost of emerging cloud service providers and with delayed project delivery.

On the surface this is a dispute between giants over the numbers. For hotel operators it is a practical question: when contracts for upstream computing capacity can be stopped just like that, what exactly are you buying when you pay for "AI customer service"?

Our view is that hotels should stop pricing AI by whether it is connected to a large model, and start accounting for how many tasks it takes off the front desk.

Why the giants hit the brakes

Two phrases in the report matter: costs too high, delivery delayed. Microsoft had previously signed large-scale contracts with emerging cloud companies including CoreWeave, Nscale and Nebius to lease computing resources for its own AI business. Re-examining those arrangements shows one thing: AI does not create value automatically once a contract is signed. After computing capacity is rented, someone has to turn it into something that does work.

The lesson for the hotel industry runs in the opposite direction. You do not need to care who is right between Microsoft and Oracle. You need to care whether the AI you have paid for is also stuck in a state where the resources are paid for but the work has not been connected.

Many hotels approach AI customer service like this: first buy a chatbot, connect it to a WeChat official account. It can answer "what time is checkout", but when a guest asks "the air conditioning in my room isn't cooling", it hands over to a human. The front desk takes just as many calls as before, only now there is an extra back end to maintain.

That is not implementation. That is one more system.

Where hotels actually get stuck

The most draining part of a high-end hotel front desk is not check-in. It is the small requests that cannot be scheduled: room status enquiries, extra beds, room changes, deposits, invoices, breakfast times. None is difficult on its own. The difficulty is that they arrive simultaneously through three channels — the in-room tablet, the guest's WeChat and the WeChat official account.

The traditional approach is to put one person on each channel, or to pull the person on one channel away to do something else, at which point the other two start to back up. Guests do not perceive that "we are busy". They perceive that "nobody is helping me".

According to the public introduction on the SHEYU AIHOTEL website, its three-channel AI customer service puts the in-room tablet, WeChat mini-program and WeChat official account into a single conversation. When a guest switches from the room tablet to a phone, the AI remembers the preceding context. Back-office staff handle transferred work orders rather than starting from scratch to understand the request. In the same system, room status, reservations, check-in, room-folio posting and settlement, and deposit management are tracked through the full workflow.

Note the boundary here: what it does is merge the three channels into one conversation and route work orders to the person who should handle them. It does not decide for the hotel whether a guest should be upgraded, and it does not apologise on the front desk's behalf. Judgements involving compensation or characterising a complaint still require a person.

How to redo the calculation

Back to the Microsoft story. The giant is re-examining cloud contracts because it found that rented computing capacity does not automatically equal business results. When a hotel examines AI customer service, the logic is identical.

You can ask any supplier three questions. When a guest switches from the tablet to a phone, does the conversation break? Does back-office staff receive the raw chat log, or a work order that states the room number? When the AI changes room status, does that flow through the same system as reservations and the room folio?

The first two questions determine whether the AI is genuinely working at the front desk. The third determines whether it becomes a new data silo.

According to its website, SHEYU AIHOTEL is aimed at high-end hotels with several hundred rooms and is positioned as a cloud PMS that integrates rooms, reservations, check-in, pricing and AI customer service. That integration comes with conditions: it suits hotels whose room count has reached a certain scale and whose front desks are genuinely overwhelmed by enquiries across multiple channels. For a small guesthouse with a few dozen rooms, where the front desk can remember every guest, the marginal benefit of installing this system is limited. That should be stated clearly.

The tightness of upstream computing contracts is decided by the giants. The efficiency with which a front desk handles one work order after another is the variable hotels control themselves.

The most practical meaning of Microsoft ending the contract is not spectacle for hotel owners. It is a reminder that the acceptance standard for AI procurement should change from "is it connected to a large model" to "how many calls did the front desk avoid taking from scratch today". The latter can be calculated. The former cannot.

閱讀繁體中文版 →