~8,900
Guest conversations
Across nine hotels and two channels
Customer case study · August 2026 · 9 properties, Germany
One month in the guest inbox
Nine budget hotels, ~8,900 guest conversations, and almost no one to answer them. McDreams runs one of Germany’s leanest hotel operations — premium sleep at budget rates, with deliberately small teams. In August 2026, D3x handled seven out of every ten replies those guests received, at an average of 4.8 seconds each.
71.3%
Replies by D3x
~20,500 of ~28,700 total replies
4.8s
Avg. time to first reply
vs 12 min 02 s for a human agent
57.6%
Fully contained
Closed with no human involved or asked for
~20,500 replies by D3x · ~8,300 by the McDreams team
August 2026 · Like Magic guest app + website chat · Volumes rounded up; percentages exact
The brand
A hotel group built to run without a front desk
McDreams has been refining the same idea since 2009: strip out everything a guest doesn’t sleep in, keep the box-spring bed, and pass the saving on. Rooms start around €30. Check-in is digital, the room key lives in an app, and there is no restaurant, no concierge and no night shift standing by.
That model works right up until the guest has a question. A lean team is the whole economics of the brand — but a guest who can’t get an answer at 23:40 is a guest who writes a one-star review about a door code. Across nine houses, that added up to more than 26,000 inbound guest messages in a single month.
2009
Founded — family-owned and family-run
9
Hotels across Germany in this study
>200k
Guests a year
~€30
Space- and staff-efficient rooms from
Digital check-in
Guests arrive, check in digitally, and go straight to their room. No front-desk queue. No overnight receptionist.
App keys
Room access lives in the phone — so messaging is the primary support channel when something goes wrong.
No night desk
When a guest needs help at 2am, there is no one at reception. The inbox has to answer — or the guest is stuck.
“D3x took us from answering guest messages to optimizing how they are handled. Two out of three conversations now resolve automatically, response times dropped from minutes to seconds, and our teams focus on the guests who actually need them.”
Channels
Two channels, one AI
D3x sits on both of the places a McDreams guest actually writes from — the guest app they check in with, and the chat on the public website. Same intelligence layer, two very different conversations.
Guests mid-stay
Like Magic guest app
Booked guests inside the Like Magic guest journey: check-in problems, door codes, late arrivals, invoices, parking, luggage. High volume, high stakes, and the channel where a slow answer costs a review.
- Conversations
- ~8,600
- Messages exchanged
- ~52,800
- Replies by D3x
- 70.1%
- Fully contained
- 56.7%
People deciding whether to book
McDreams website
Anonymous visitors asking about rates, parking, pets, cancellation and availability — the questions that decide a direct booking. Lower volume, and almost entirely self-service.
- Conversations
- ~310
- Messages exchanged
- ~2,400
- Replies by D3x
- 99.7%
- Fully contained
- 83.9%
On the website channel, four of ~1,200 replies came from a human. The AI is effectively running that channel unaccompanied.
Containment
What “handled” actually means
Deflection numbers are easy to flatter. This is the strict version: a conversation only counts as contained if no human sent a message and the guest never asked for one.
100%
~8,900
Guest conversations started
All conversations opened across both channels in August 2026
62.8%
~5,600
Handled by AI alone
No human agent sent a single message
57.6%
~5,100
Fully contained by AI
No human message, and the guest never asked for one either
Of the ~3,000 conversations where a guest did ask for a human (33.9%), the request itself is a healthy signal — D3x hands over rather than looping. The number worth watching is the ~470 where nobody on the team picked the handover up.
Speed
The gap is not minutes. It is two orders of magnitude.
Average time to reply on the Like Magic channel, August 2026. D3x answered roughly 150× faster than the human average — and did so at 02:00 as readily as at 14:00.
4.8s
D3x average first reply
Same channel, same month
12m 02s
Human average first reply
Like Magic guest-app channel
~150×
Faster first reply
D3x vs human average
AI vs human outcomes
The conversations D3x finished alone were resolved more often than the ones a human joined.
This is not evidence that the AI is better at hospitality than the McDreams team. It is evidence of what gets escalated. The conversations a human joins are, by construction, the hard residue.
+12.7 pts
Higher resolution rate for AI-only conversations
+0.38
Higher guest satisfaction score for AI-only conversations
AI alone
57.3%
Resolution · satisfaction 6.43
AI plus a human agent
44.7%
Resolution · satisfaction 6.05
Share of conversations where the guest’s problem was resolved · ~5,600 AI-only conversations vs. ~3,300 human-assisted, August 2026
What the numbers do show is that the routing works. The straightforward 58% never reaches a person at all, and the team’s attention lands where the outcome is genuinely uncertain. Satisfaction on AI-only conversations sits at 6.43 against 6.05 when a human is pulled in — a lean team spending its hours on the cases that actually need judgement.
Capacity
What that buys a nine-hotel team
≈ 340 hours
of front-of-house messaging work absorbed in a single month — roughly two full-time equivalents, spread across nine properties, without adding a night shift or an outsourced contact centre.
- ~5,100 fully contained conversations × 4 minutes of agent handling time each ÷ 60 ≈ 340 hours
- At a 160-hour month that is 2.1 FTE
- The 4-minute assumption is McDreams-specific and should be replaced with the group’s own average handling time before this figure is quoted externally
Methodology
How these numbers were produced
Every figure on this page comes from D3x platform reporting for 1–31 August 2026, covering all nine McDreams properties across the Like Magic guest-app channel and the McDreams website chat. Conversation and message counts are logged directly, then rounded up for display; percentages, response times, and satisfaction scores are shown exactly. Satisfaction and resolution rates are estimated by sentiment analysis of the conversation text rather than collected by survey, so they are best read as a comparison between groups, not as absolute scores. The capacity figure is a model, and its assumption is stated in full alongside it.
Next step
See how it would work on your stack
Chatbots answer. D3x resolves. The orchestration and intelligence layer that sits above a hotel’s existing stack — reading and writing to the systems the team already runs on, and finishing guest requests end to end.