Most guides to hotel chatbots answer the wrong question. They explain whether AI can hold a conversation, when the deciding issue for a hotel is what happens after the AI has understood the guest: whether the check-in is completed, the ID verified, the deposit taken, the door code issued, the guest registered with the authorities, and the broken air conditioning assigned to somebody who can fix it.
This guide covers what AI systems connected to hotel operations can actually do, what still needs your staff, which integration questions to ask before you buy, and how to tell a basic FAQ bot from a system that executes.
Basic FAQ chatbots versus AI connected to operations
A scripted or FAQ chatbot answers questions from a set of predefined replies. It handles opening hours, wifi, parking, pool times, pet policy. That work is genuinely useful: it removes repetitive messages from the front desk and answers guests at 23:00.
Where it stops is where most guest requests actually need something done. “Can I check in early?” is not a question, it is a request that has a price, an availability check and a confirmation. “There is water on the bathroom floor” is not a question either; it is a task with a priority and an owner.
The distinction that matters is not rule-based versus AI. It is answering versus executing. An AI system connected to your property management system and your access providers can read the reservation, complete the check-in, collect the deposit, issue the code, prepare the guest registration and create the task. Whether yours does any of that depends on the integration, not on the quality of the language model.
Seven workflows worth automating, and what each one depends on
1. Guest messaging around the clock, in the guest’s language. Repetitive questions in the pre-stay, in-stay and post-stay phases are answered automatically, with the language detected from the guest’s own message. The limit is your content: an AI only knows the policies, prices and contacts you load into it. Budget for maintaining that knowledge.
2. Image and voice understanding. A photo of a thermostat error code or a voice note about a late arrival can be read, classified and converted into the right conversation with the attachment attached, rather than a message someone has to interpret later. Ambiguous images still need review, so decide with your team which incident types always go to a person.
3. Online check-in with ID verification. Data and documents are collected before arrival and identity is verified by scanning, so reception is not doing data entry while a guest waits. What can be automated here depends on the PMS and access provider connected.
4. Authority guest reporting. Where the law requires accommodation providers to report guest data, the data collected at check-in can feed the submission instead of being typed twice. UpMarket handles this for Spain (Mossos d’Esquadra in Catalonia, SES nationally), Portugal and Italy. The duty’s scope depends on the property type and the rules that apply to it, so confirm the details for your own property before relying on any tool, ours included.
5. Deposits and payments. A deposit is a process with states — authorised, consumed, released — not a message. Asking for it during check-in, and releasing it when the stay closes correctly, removes a task nobody enjoys chasing. Payment gateways, amounts and timing are your decisions, and the local tax rules are your responsibility.
6. Access. Door codes or digital keys can be issued when the check-in completes. This one depends entirely on your lock or access provider: if it is not compatible, or does not expose the necessary actions, no amount of AI quality will bridge that gap.
7. Escalation and assignment. When a conversation needs a person, it should become a task with context, an owner and a priority, visible to the next shift, with internal notes the guest never sees. Escalation rules and on-call rotas are yours to decide; automation does not remove the need for someone to be reachable.
What still requires your team
No honest demo promises autonomous confirmation of a stay extension, identical write-back to every PMS, workforce scheduling, or overnight repairs. Extension requests still need price and availability authority, connector depth varies by PMS, and staffing is a different category of software. Complaints that require compensation, relocation or an apology need judgement and authority. Deciding what not to delegate is part of the design.
Integration questions that separate vendors
A logo wall tells you nothing. Ask for the specifics of your system:
- Which reservation data flows in from my PMS, and how often — created, updated, cancelled, checked in, checked out, no-show?
- Which actions can your system take back to my PMS, and which are read-only?
- What happens to a conversation when the PMS is unreachable or the connection breaks?
- Is the AI answering from my reservation data or from generic content?
- Which access or lock providers are supported end to end, and which only partially?
- Can someone take over a conversation mid-thread, and does the AI stop replying while they hold it?
- How are internal notes kept separate from guest-visible messages?
- What do you need from us before go-live: policies, prices, contacts, escalation rules, rotas?
- How do you handle a wrong answer, and how do we correct it?
- What does the setup and onboarding actually involve, and who maintains the integration afterwards?
Vendors document different levels of this publicly. Chatlyn names eleven PMS connectors and states that reservation events flow into its chat, without documenting write-back actions. HiJiffy documents one PMS in depth — five integration types, upsell orders written back to the bill, check-in links and digital keys — and advertises integration tiers (real-time, scheduled or CSV) for the rest without saying which PMS gets which. Canary says it has two-way integrations with Opera, RoomMaster, StayNTouch, Mews, Cloudbeds and others, without detailing what each one supports. Duve names fourteen PMS or channel managers and nine lock brands. None of them publishes a matrix of which actions each connector supports, which is the question to put on the call. Treat all of it as a starting point for the call, not as the answer. UpMarket’s own partner directory is public and deserves the same reading: what the entry says the connection supports matters more than the logo.
Commercial benefits, stated honestly
Automating the operational parts of the guest journey changes three things: the front desk spends less time on data entry and repetitive messages, guests get an answer or a completed task at the hour they actually ask, and extras are proposed inside a conversation where the guest has already shown interest. Upsell revenue captured this way is real but modest and depends on your offer; anyone quoting a fixed uplift percentage without showing you the underlying data is guessing.
What to prepare before you switch it on
Write down your policies, prices, cancellation terms, escalation rules and who is on call, in the languages your guests use. The quality of that document will show up in the quality of the answers. Then run a pilot on one property or one channel, review the conversations weekly, and correct the content rather than blaming the AI.
FAQ
Is an AI chatbot the same as a virtual concierge? A virtual concierge is the guest-facing part; what separates platforms is whether the conversation is connected to the check-in, the deposit, the access and the reporting behind it.
Do we still need 24/7 staff? No system replaces a person for urgent incidents or decisions. Automation reduces the volume of routine work, so the person you do have can handle what matters.
Will it work with our PMS? Ask for the specifics of your system, and test them in a demo with real questions, not a slide.
Can it handle several languages at once? Language detection and multilingual replies are standard in this category. Ask how the team reads a conversation held in another language.
How long does setup take? It depends on your PMS, your access provider and how much content you need to write. Ask for a sequence with named steps, not a duration.
See the workflows, not the marketing
The useful part of a demo is the part after the AI understands the guest. Try UpMarket’s live demo and ask it for the access code before check-in, an early check-in, a paid extra and an urgent fault — then see what it does with each. You can also create an account or see the plans.
For more on the operational layer, read how guest communication software connects to your PMS, how escalation rules decide whether 24/7 WhatsApp messaging helps, why stopping at automated check-in leaves the rest of the journey manual and how we compare with Enso Connect.
