At C-Hotels Andromeda in Ostend, a guest message gets answered in a median of under ten seconds. At the same hotel, when a message does reach a person, the median staff reply runs over an hour.
Same building. Same guest. Same question about parking. Two completely different waits.
Hotel response time is usually filed under service standards, somewhere between turndown and the breakfast buffet. It belongs in the revenue conversation instead. The gap between a reply that lands in seconds and one that lands in forty minutes is the gap between a traveller who books you and a traveller who books the hotel two streets over.
Two clocks run in the same building
Across the three properties AskWhisper publishes numbers for, the automated clock reads the same everywhere: a median answer in under ten seconds, with the 90th percentile under twenty.
The human clock reads very differently, and both properties that publish it are honest about the number.
| Property | Median answer, automated | 90th percentile | Median when a person takes it |
|---|---|---|---|
| C-Hotels Andromeda, 114 rooms, Ostend | Under 10 seconds | Under 20 seconds | Over an hour |
| Yalo Ghent, 92 rooms, Ghent | Under 10 seconds | Under 20 seconds | Over 30 minutes |
| C-Hotels Silt, 73 rooms, Middelkerke | Under 10 seconds | Under 20 seconds | Not published separately |
Those staff figures are not a criticism of the teams. Andromeda answers more than 6,500 of 7,500+ messages with nobody involved, so the few hundred that do reach a person are, by definition, the hard ones. Yalo is the same shape: more than 2,400 of 2,500+ messages over ninety days handled without staff. At Silt, over the last ninety days the AI answered 4,000+ guest messages and the team answered fewer than 25.
An hour is what a good front desk does with the questions that actually need judgement, while also checking people in.
What travellers say they will tolerate
The industry's most-quoted benchmark on this comes from research ALICE published in October 2016, an online survey of more than 400 frequent travellers in the US. It found guests felt entitled to complain if a hotel took longer than about twelve minutes on a text message, twenty-six on email, and twenty-seven on social.
Two things about that number. It was commissioned by a hotel technology vendor, and it is ten years old. It predates WhatsApp becoming the default channel for European travellers entirely.
Twelve minutes was the tolerance of someone whose other apps were slower than yours are now. Nobody has published a better figure since, which tells you how little attention this metric gets.
What happens in the gap
A traveller who does not get an answer does not wait politely. They do one of three things.
They ask somebody else. Before booking, that means the OTA listing, a competitor's chat widget, or an AI assistant that will happily answer for you using whatever it can find. The enquiry does not pause while you get to it.
They book elsewhere. A question asked at the point of deciding is a buying signal. Left open, it resolves as a no.
Or they walk to the desk. This is the one hoteliers underestimate. Every unanswered message becomes a person standing in front of your receptionist during check-in, which is the single worst moment to absorb an extra conversation.
The questions doing this volume are not complicated. Andromeda's case page names the ones that come all day at a coastal property: parking, breakfast, wellness hours. Add check-in time and wifi and you have most of it. Parking alone at Andromeda produced more than 500 reservations placed by guests themselves, with no staff involvement at all.
None of those five questions needs a human. All five of them, unanswered, cost you something.
Speed is a conversion metric before it is a service metric
Before the booking, the reply either arrives while the traveller is still deciding or it arrives after they have decided. There is no third option.
After the booking, speed does something different: it opens a conversation that would otherwise never exist. Silt is the clearest evidence of this. Before May, under 6% of arriving bookings ever received a message. Now seven in ten arrivals get one, and more than half of them write back.
More than half. That is not a service metric, that is a channel that did not exist before May. It is also where the €24,000+ of upsell revenue at Silt comes from, because breakfast and parking can only be sold to a guest you are actually talking to.
The order matters. Reach opens the conversation, speed keeps it alive. A guest who writes back and waits forty minutes for an answer does not write back a second time.
The 90th percentile is the number that matters, not the average
If you only track one figure, track the slow tail.
Averages are generous to hotels in a way that hides the actual failure. A property that answers ninety messages instantly and leaves ten sitting overnight still posts a respectable average. The ten it left are the ones that cost a booking: the 2am message in German, the Sunday enquiry, the question that arrived while the duty manager was dealing with a broken lift.
Medians are better. They tell you what a typical traveller experiences rather than what your arithmetic experiences.
The 90th percentile is better still, because it tells you what your worst tenth of travellers experience. Across Silt, Andromeda and Yalo, that number sits under twenty seconds. Not because every message is easy, but because the same system answers at 3am in German as at 3pm in Dutch.
The average tells you how your hotel performs on a good Tuesday. The 90th percentile tells you how it performs at the moment a traveller decides whether you are worth the risk.
Some messages should not be answered in four seconds
This is where the speed argument has a limit, and pretending otherwise would be dishonest.
A complaint answered in four seconds reads as dismissal. The guest who writes "the room next door was loud until 3am and I have a meeting at nine" does not want an instant reply. They want evidence that a person read it, felt something about it, and can do something about it. Speed there actively damages you.
The same applies to any request where the answer costs the hotel money or requires a discretionary yes. A late checkout on a sold-out Saturday, a rate adjustment, a booking change outside policy, a proposal for twelve rooms. Nobody wants a fast maybe. They want a slow yes from somebody with the authority to give it.
The handover belongs in three places:
- Anything emotional. Complaints, disappointments, anything where the hotel got something wrong.
- Anything discretionary. Requests where the answer depends on availability, revenue, or a judgement call about this specific guest.
- Anything unknown. When the knowledge base does not have the answer, the correct behaviour is to say so and pass it on, not to improvise.
Everything else is a factual question with a factual answer, asked forty times a week, and there is no service value in making someone wait for it.
The version that works is both: instant acknowledgement, human resolution. The guest knows within seconds that the message landed. A person picks it up with the context already attached. We draw the full line in what to automate and what to keep human.
How to measure your own response time this week
Three checks, none of which require buying anything.
1. Calculate the median, not the average. Pull your last hundred inbound messages across every channel. Sort by time-to-first-reply. Take the fiftieth. Then take the ninetieth and look at it honestly, because that is the number your reviews are written about.
2. Split it by channel. WhatsApp, webchat, email, Instagram, the booking engine enquiry form, the phone that goes to voicemail. Most hotels discover one channel nobody actually owns. It is usually the one a traveller uses before they have booked anything.
3. Split it by hour. Map your reply times against a 24-hour clock. The shape tells you everything: a clean line from 08:00 to 22:00 and then a cliff. Every message in that cliff waited until morning, and international travellers live in it.
If all three of those come back clean, you do not have a response time problem. Most properties find they have one specific hour, one specific channel, or one specific question eating the difference. If it's WhatsApp, our WhatsApp automation playbook covers the setup.
Want to know what your own two clocks read?
Pull your last hundred messages and find the ninetieth slowest reply. If that number is measured in hours rather than seconds, see what AI replies look like on your property. No demo call required. Andromeda set itself up without a sales call.
Turn lookers into bookers. Just Ask Whisper.
