AI voice & chat concierge · Prepared for the Moudira Collection · September 2026

Your guest is awake
in New York.
Your desk is asleep
in Luxor.

Syntra builds and runs the AI voice and chat systems that carry a property's front line when nobody is there to answer. In production since March 2026 on a 46-cabin Nile cruise ship and, since August, on a luxury dahabiya. This document was written for the Moudira Collection specifically, and every number in it is checkable.

Luxor --:--
New York --:--
Reservations Closed
Concierge Online
Prepared by
Seif Said · Syntra · Egypt
Reasoning layer
Anthropic Claude · Anthropic-certified engineers
Status of the numbers
Verified and modelled, kept separate
Syntra · The Moudira Collection · September 2026Scroll
Why this document exists

Most vendors will show you a hotel product.
You are not buying for a hotel.

The Moudira Collection is seven properties in four different booking shapes, five of which are not hotels at all, sold from one reservations desk in a time zone that runs ahead of every market that books them.

Syntra is the only vendor in this process that already runs a concierge on the Nile: on a 46-cabin cruise ship since March, and on a Small Luxury Hotels dahabiya since August, both carrying live guest traffic today. The reasoning layer is Anthropic's Claude, our engineers are Anthropic-certified, and the single behaviour our operators chose us on is what the system does when it does not know the answer: it says so, in the guest's language, and routes to reservations. It never guesses a price or a date.

What follows separates what our live systems have recorded from what we have modelled, and ends with seven questions we think you should ask everybody in this process, including us.

Mar ’26
In production since
Continuously, without a rollback
2
Properties live
A Nile cruise ship and a dahabiya. A ten-property hotel group in conversation.
1,650
Guest questions answered
Across 600 live conversations on the flagship
7
Languages
AR · EN · DE · FR · IT · ES · PT
Contents
§01The Collection: seven properties, four questions §05The model layer: Anthropic-certified §02The fit: proven on this exact stretch of river §06The evaluation: seven questions for every vendor §03Evidence: verified and modelled, kept separate §07Regulated: working where somebody checks §04Engineering: the eight subsystems §08Proposal: what we would do for the Collection
§01The Collection

Seven properties.
Four completely different questions.

A Relais & Châteaux hotel, private villas with butlers and chefs, a house on its own stretch of the Nile, a villa in an oasis, Art Deco apartments in Cairo, and two dahabiyas that sail. All sold from one desk, two hours ahead of London and seven ahead of New York.

CAIRO ImmobiliaSERVICED APARTMENTS FAYOUMBeit Fayoum14 ROOMS LUXOR · WEST BANK Al Moudira Hotel Al Moudira Villas Beit El Nil 54 SUITES · 6 VILLAS · 12 SUITES ESNA EDFU KOM OMBO ASWAN SET NEFRU · NILE CANOPUS · 3 TO 7 NIGHTS Two dahabiyas, sailing
The Collection along the river. Copper marks a Moudira property. Not to scale. Counts from moudira.com and moudiracollection.com, September 2026.
Al Moudira Hotel
Luxor · West Bank54 suites · 10 courtyards

Six suite categories, Egypt's first Relais & Châteaux member. The volume asset; its enquiries arrive overnight from Europe and North America.

Set Nefru
River Nile · Luxor to AswanDahabiya · 5 cabins · private charter

A ninety-year-old sailing boat, relaunched in 2023, taken whole for up to eight guests on three to seven-night voyages. The charter rule our live systems already enforce.

Nile Canopus
River Nile · Luxor to AswanDahabiya · cabins sold individually

Contemporary dahabiya, cabins booked one at a time. The cabin-and-itinerary logic our flagship deployment has run since March.

Al Moudira Villas
On the hotel estate · Luxor6 villas

Butler, private chef and concierge. The enquiry is a conversation about a whole stay, not a room rate. Category tools have no vocabulary for this.

Beit El Nil
Luxor · private Nile frontage12 suites

Sold by the suite and taken whole. Buyout logic is exactly the rule set we built for a dahabiya operator, and the rule most chatbots quote wrongly.

Beit Fayoum
Tunis village · Fayoum Oasis14 rooms

A different governorate, a different transfer question, a different season. Guests ask how to get there before they ask what it costs.

Immobilia
Downtown Cairo1 to 3 bedrooms

Serviced apartments in Cairo's first skyscraper, a 1930s Art Deco landmark. Length of stay, not nights. Nothing in a hotel PMS models this correctly.

§02The Fit

We have already done
the hardest part of this.

Both live deployments
sail this stretch of river

Not hotels with a river view: vessels between Luxor and Aswan, fixed itineraries, seasonal pauses, cabin categories with occupancy rules, charters, and a booking conversation that is half logistics and half persuasion. Historia is a 46-cabin Nile cruise ship and has carried guest traffic without a rollback since March. Storia is a dahabiya and a member of Small Luxury Hotels of the World, which means its knowledge base was built and verified to the standard a soft brand expects.

Every other vendor in this process will be learning the dahabiya from your team during onboarding. We would be arriving with it built. The parts that take a general vendor a quarter to get wrong and then right are already written, tested and in production: seasonal sailing pauses, the difference between a cabin and a charter, multi-night totals that do not lie, and what the system says when the booking system is briefly unreachable.

01
Nile logic, already live

Cabins, charters, itinerary lengths, occupancy rules and the summer pause are production behaviour, not a roadmap.

02
The Relais & Châteaux languages

French, German, Italian, Spanish, Portuguese, Arabic and English. Your core European source markets are covered on day one.

03
Egypt, on the ground

Onboarding is people in a room with your reservations team, not a video call across eight time zones.

The argument no single-property tool can make

One concierge, seven doors.

A guest asking about four nights on Nile Canopus is, in the same conversation, asked whether they would like two nights at Al Moudira before they sail, because the boat leaves from Luxor and the hotel is on the West Bank. A guest enquiring about Immobilia in Cairo is offered the oasis. Seven reservations desks cannot do that at nine in the evening New York time. One concierge that holds the whole Collection in a single knowledge base does it in every conversation, and the telemetry tells you afterwards which pairings guests actually reach for.

§02A dahabiya enquiry

Watch the charter rule
get answered correctly.

Set Nefru is taken whole. Nile Canopus sells cabins one at a time. A couple asking about the wrong boat is the most common enquiry a Nile operator gets, and the one most chatbots answer wrongly. This is the exchange as our live systems handle it.

Simulated · illustrative, not a real transcript

Website chat · EnglishGuest · New York · 20:41
Guest
We're thinking about a Nile trip in February, just the two of us. Can a couple book a cabin on Set Nefru, or do you have to take the whole boat?
Concierge
Set Nefru is taken as a whole boat: five cabins, up to eight guests, so it suits a family or a group sailing privately. For two of you there is Nile Canopus, the Collection's contemporary dahabiya, where cabins are sold individually on the same Luxor to Aswan route. Would you like me to look at Nile Canopus for February, or hold Set Nefru privately?
Guest
Nile Canopus then. Around the 14th, four nights maybe.
→ Live availability checked against the booking system: Nile Canopus, one cabin, 14 to 18 February
Concierge
There is a cabin on the sailing that leaves Luxor on 14 February, four nights, arriving in Aswan on the 18th. Would you like to see the cabin first, or the route?
Guest
The cabin.
→ Cabin gallery opens inside the conversation. Guest swipes through four photographs.
Guest
Lovely. What would that come to for the two of us?
→ Full four-night total for two resolved from the booking system at the February rate, quoted as one figure with what it includes, never per night. Had the booking system been unreachable here, the concierge says so and routes to reservations. It does not estimate.
Guest
Great. Could we add a couple of nights at Al Moudira before we sail?
Concierge
Yes, and that is the usual way to do it: the boat departs from Luxor and the hotel is on the West Bank, so the transfer is short. I will put the nights of 12 and 13 February into the same request so reservations sees the whole trip together. Shall I send it to them to confirm?
Guest
Yes please.
→ Booking inquiry created: Nile Canopus, one cabin, 14 to 18 February, two guests; Al Moudira, 12 to 13 February. Reservations notified with the full transcript attached. Guest receives a written summary. The desk sees it at 08:00 in Luxor.
ElapsedUnder four minutes
Human involvementNone until the desk opens
Runs the same inFR · DE · IT · ES · PT · AR
§03Evidence

What it has
actually done.

We separate what the live systems recorded from what we have modelled. The first set is countable. The second is our estimate and is labelled as such, because a buyer should be able to check us and find that the numbers hold.

● Verified · flagship deployment · March to August 2026
1,650
Guest questions answered
Counted from live transcripts, test sessions excluded
600
Live conversations
Each a complete, resolved exchange
230
Booking inquiries generated
Handed to reservations with the transcript attached
2.75
Questions per conversation
Guests keep asking once they know they will be answered
100%
Resolved in full
No bounces, no dead ends
0
Rollbacks
Six releases into a property that was actively selling
Historia
Live · March 2026

Nile cruise ship, 46 cabins, 88 guests, Luxor to Aswan. The flagship, the source of every verified figure above, and still the longest running record we have.

Storia the Dahabiya
Live · August 2026

The AND Group's dahabiya, a member of Small Luxury Hotels of the World. Charter-exclusive periods, an eclipse sailing, one open date a month. Its first weeks of traffic are not yet in the figures above.

Azur Hospitality
In conversation

An Egyptian hotel group of ten properties. A bespoke pitch was built for the account and the conversation is live. Not contracted, and we will not count it until it is.

Nothing on this list came from a marketplace or an agency introduction. The flagship went live, the numbers came back, and the same operator group put its dahabiya on the same stack.

§03Evidence · what the telemetry gives back
Which departure months guests ask aboutFlagship · 262 requests · once per conversation
13
68
35
43
13
8
7
8
13
19
21
14
JanFebMarAprMayJunJulAugSepOctNovDec
October to April · 213 of 262 (81%)May to September

That chart is not a marketing statistic. It is a live count of which month guests actually asked to travel in, taken from conversations, before anybody booked anything. The West Bank runs on the same cool-season calendar as the river, which means the Collection's own version of this chart would tell your commercial team where to put rate and where to put spend, on a signal that currently disappears the moment a guest closes the browser.

The forward book is the part operators tend not to expect. On the flagship, the single strongest month on the board was November of the following year, with October and December behind it. Guests were asking about departures well over a year out, as far as November 2027. Nobody was capturing that before the concierge went live, because there was nobody awake to be asked.

Nov 202618Strongest forward month on the book
Oct 202617Season opening
Dec 202614Holiday sailings
Jan 202712New-year departures
Mar 202711Nineteen months ahead of the ask
○ Modelled · our estimate, not measured
88%
Zero human handoff
Conversations resolved without a person
<2s
Time to first answer
Voice and chat
60h
Staff time freed, monthly
Per property: 20 to 25% of the support and sales load
34%
Upsell attach rate
Excursions, dining, a second property
“You were extremely helpful... is there any way I can save your hard work?”
Guest planning February 2027 · chat
“The luxury cabin is perfect for my daughter and I.”
Guest booking the 23 October sailing · chat
“Está perfecto, gracias.”
Guest served end to end in Spanish · chat

Verbatim guest messages from live conversations. Internal test sessions excluded; names and contact details withheld.

§04Engineering

A year of engineering
before the first guest said hello.

The concierge is not a wrapper around a chat model. It is a real-time system with eight subsystems that each had to be right before a paying property would put its name on it, and the hardest of them is the one that decides when not to answer.

01
Real-time voice pipeline

Streaming speech recognition into reasoning into speech synthesis, with interruption handling so a guest can cut in mid sentence the way they would with a person. Voice and text run on one brain, so a conversation that starts on voice can finish in chat.

05
In-conversation media

Ask to see a suite and a swipeable gallery appears inside the conversation. Ask about an itinerary and an interactive map of the Nile opens: tap Luxor, Esna, Edfu, Kom Ombo or Aswan for day by day schedules. On voice and chat both.

02
Per-property knowledge compilation

Every deployment gets its own knowledge base, built with the operator and verified fact by fact: rooms or cabins, itineraries, transfers, dietary handling, accessibility. Prices and live availability are deliberately excluded, so the concierge cannot recite a stale rate.

06
Guest memory

Returning guests are recognised and welcomed back by name, with the concierge picking up where the last conversation ended: dates, preferences and all. Nobody repeats themselves.

03
Live tool-calling into the property's systems

Availability, quotes, multi-night totals and seasonal rate transitions are resolved against the real booking system during the conversation, not from a cached table.

07
Telemetry and operator dashboard

Every conversation is timestamped, geolocated, language-tagged and categorised. The operator sees which questions are being asked, which markets are asking, and which departure dates guests keep reaching for.

04
Honest-failure design

When an upstream system is briefly unreachable, the concierge says so and routes the guest to reservations. It never guesses a price or a date. This is the single most important behaviour in the product, and it is why operators let it speak to their guests unsupervised.

08
Omnichannel surfaces

Website voice and website chat are in production today. Telephone and WhatsApp run on the same brain, the same knowledge base and the same transcript history, and are switched on per property.

Five releases in thirty days, into a live client
2.0 · 11 Jun
Reliability
Honest failure when the booking system is unreachable; seasonal awareness through the summer pause
2.1 · 4 Jul
Returning guests & team access
Guest memory across sessions; dashboard sign-in for the operator's team
2.2 · 6 Jul
Pricing accuracy
Full-stay totals, never per night; excursion pricing follows seasonal rates through the October transition
2.3 · 9 Jul
Photo galleries
Cabins, suites, sun deck and dining, on voice and chat
2.4 · 10 Jul
Interactive itineraries
A live map of the Nile, all four itineraries day by day

Each went out with a written changelog the operator could read, while guests were mid conversation. Version 2.5, returning-guest recognition, followed on 14 July.

§05The model layer

Anthropic-certified,
and why that is the part that matters.

The reasoning layer is Claude

Syntra's reasoning runs on Anthropic's Claude, and our engineers hold Anthropic certification. We did not arrive there by default. We chose it, and stayed on it, because of a single behaviour: what a model does at the edge of its knowledge.

A concierge that invents a plausible answer is worse than no concierge at all, because the damage is not a bad conversation, it is a guest who arrives expecting something the property never offered. Every operator in luxury hospitality knows this instinctively, which is why so many have looked at this category and quietly kept answering the emails themselves. Ours let the system speak unsupervised because it declines cleanly, in the guest's language and the property's voice, without sounding like a failure.

What the certification covers

Building production systems on Anthropic's platform: tool use, retrieval, evaluation, and the safety behaviour that governs what a deployed agent is permitted to say.

What it buys the Collection

The people writing your concierge's behaviour are certified on the model it runs on, rather than prompting a general purpose API and hoping. The refusal behaviour is engineered, tested and versioned.

What it does not mean

It does not mean Anthropic endorses Syntra, and we will not imply that it does. A partner who overstates a certification will overstate an availability quote.

§05Stated plainly

What it will not do.

It will not sound like a call centre.

Al Moudira was built by hand over twenty-five years and staffed from the villages around it. The concierge's register, phrasing and what it may volunteer are written with you and reviewed by you before a guest hears a word.

It will not invent.

If an answer is not in the verified knowledge base or the live booking system, it says so and routes to reservations. This is engineered, tested and versioned, not hoped for.

It will not replace your people.

It is the front line at the hours and in the languages the desk cannot cover. Reservations still confirms every booking, and a guest who asks for a person is handed to reservations with the transcript attached, so nobody starts again.

It will not keep your guests' conversations from you.

Every transcript and every inquiry is the property's record, visible in your dashboard with access for your own team, and exportable.

§06The evaluation

Seven questions to ask every vendor in this process.
Including us.

You are looking at other people, and you should be. These are the seven questions that actually separate the products in this category, with our own answers written underneath so you can hold us to them in the same room as everybody else. Any vendor who cannot answer all seven in one meeting is asking you to find out the answers during your season.

Q1
When a guest asks for a price, does the system quote from the live booking system, or from something you loaded last month?

Most of this category reads availability and captures a lead. Quoting is where they stop, because a wrong price is a liability and the safe product is one that hands you an email address.

SyntraAvailability, multi-night totals and seasonal rate transitions are resolved against the real system during the conversation. Version 2.2 exists specifically because a multi-night quote must always show the full stay total, never a per-night figure a guest can misread.

Q2
What does it do when it does not know, and can you show me that happening?

Ask for a live demonstration of a failure, not of a success. Vendors in this category rarely specify hallucination behaviour in writing, and almost none will show you the system breaking on purpose.

SyntraHonest-failure design is our fourth subsystem and shipped in our foundation release. We will disconnect the booking system in front of you and let you watch what the guest sees.

Q3
Is it one product on voice and chat, or two products with one invoice?

The category splits in half here. Chat-first platforms have voice as an attachment with variable latency; voice-first specialists do not do the messaging well.

SyntraWebsite voice and website chat run on a single brain, a single knowledge base and a single transcript history, with telephone and WhatsApp on the same agent. A guest can start on voice and finish in chat without repeating themselves.

Q4
Does your integration list name the systems the Collection actually runs?

“PMS integration” is not a promise, it is a category. The published integration lists in this market frequently do not include the specific system a given property runs, and at least one prominent vendor is verified against a single PMS today.

SyntraWe build against whatever each property actually uses, asset by asset, and where an asset has no usable interface we tell you before the contract rather than after.

Q5
Have you deployed on anything that is not a hotel?

This is the question that should decide it. Five of your seven assets are not hotels. A five-cabin sailing boat with three to seven-night itineraries and a charter rule does not fit a room-and-rate-plan model, and neither do serviced apartments in Cairo or a villa with a private chef.

SyntraBoth of our live deployments are vessels on the Luxor to Aswan route: a 46-cabin cruise ship and a dahabiya. We are not learning this asset class on your account.

Q6
Who operates it after the signature, and what is your release record?

Most of this category sells you a platform and leaves you to configure it. Ask for a changelog.

SyntraFive documented releases in thirty days into a property that was actively selling, each with a written changelog the operator could read, shipped while guests were mid conversation. We design the system, write it, deploy it and then run it, including incidents and knowledge updates.

Q7
What is the price, in writing, with nothing bundled?

Pricing in this category is routinely bundled across message volume, connectors, channels, booking assistant, guest app, setup and AI add-ons, which makes two quotes almost impossible to compare.

SyntraTwo numbers: a one-time build, and a monthly fee per site. The pilot is priced to be an easy yes. The estate is priced on the reference the pilot produces.

§07Regulated

Why a regulator's file belongs
in a hospitality pitch.

Alongside the concierge practice, Syntra is building ARK, a retail investing application for the Egyptian market, past initial application and moving through the Financial Regulatory Authority toward its sandbox. Delivered on iOS: an Egypt Exchange section covering the EGX 30, 70 and 100, market breadth and sector performance, alongside US equities and options, brokerage and order entry, and an AI portfolio agent on the same voice-and-reasoning stack as the concierge. It is pre-launch: no cash rail, no live trading venue, no retail customers, and broker orders run against a paper environment while the submission proceeds. Behind it sits the paperwork a regulator actually reads: a system description, a statement of applicability, and a risk assessment mapped to the FRA's control set, written so that every factual claim names the file it was read out of. Where an answer is not yet known, the document says so instead of guessing.

It is here for one reason. A regulated build is proof that a team can work inside somebody else's expectations: evidence trails, data-boundary questions, failure behaviour, and documents that survive review. Relais & Châteaux membership comes with a standards audit. Those are the same questions in a different vocabulary, and we have already answered them for an audience that checks.

ARK, Syntra's retail investing app for Egypt, onboarding screen
§08Proposal

What we would actually do
for the Collection.

We would not start with seven properties, and we would be suspicious of anybody who offered to. Nor would we start with the easiest one. We would pilot on two assets at once, chosen to be as different as the Collection contains.

Phase one · pilot
Al Moudira Hotel and Nile Canopus, together

One land property and one sailing vessel. The flagship gives us the volume, the overnight traffic and the six suite categories. The dahabiya gives us cabins, itineraries and the charter rule, with Set Nefru's private-hire logic answered in the same knowledge base. If one concierge handles both correctly, the remaining assets are knowledge engineering, not a question of whether the system works.

Phase two · the river and the houses
Set Nefru, Beit El Nil, the villas

Set Nefru inherits almost everything from Nile Canopus. Beit El Nil's buyout logic is the rule set we built for a dahabiya operator, applied to a house. The villas add butler, chef and private concierge enquiries, which is a knowledge problem rather than an engineering one.

Phase three · outside Luxor
Beit Fayoum and Immobilia

Different governorates, different transfer questions, and in Cairo a length-of-stay model rather than a nightly one. These go last because they are the least like the others, not because they matter least.

The calendar · why this decision has a date on it

The season opens in October. February is the busiest enquiry month on the river: 68 of 262 departure requests on the flagship, more than May to September combined. A pilot live by December answers through the peak and produces a reference before the season ends. One that starts in January is being built through it. That is the difference between deciding the estate on evidence next spring and on a promise next autumn.

Success criteria · agreed in writing before launch

Six numbers we will be measured on.

Share of conversations resolved without a human handoff
Booking inquiries generated per week, by property
Share of conversations arriving outside desk hours
Languages served end to end
Cross-property requests: guests offered and accepting a second asset
Incorrect price or date statements. Target: zero. We will put that one in the contract.
§08How an engagement runs
01
Discovery

We learn the assets and the buyer properly. What each one sells, who signs, what guests object to.

02
Knowledge engineering

Everything the system will say, verified with you fact by fact. The step most vendors skip, and the reason ours can be trusted unsupervised.

03
Build and integrate

Connected to the real systems of record, with failure behaviour defined before launch behaviour.

04
Pilot, live and measured

Nobody signs an estate on a promise. They sign it on a reference.

05
Expand

The pilot's own numbers open the collection conversation. That is exactly how one vessel became two.

Commercial shape

A one-time build against a monthly per-site operating fee, with the pilot priced to be an easy yes and the estate priced on the reference the pilot produces. Every direct enquiry the concierge converts is one that did not go through an intermediary. Nothing on our ladder came from a marketplace or an agency introduction: the flagship went live, the numbers came back, and the same operator group put its dahabiya on the same stack. That is the motion we would run here, and it is the only kind of proof we think is worth putting in front of you.

Your desk opens at eight.
The concierge never closed.

Seif Said · Syntra · Egypt
seif.said@syntra.to
Web
syntra.to
This document
Download the PDF