AI receptionist for a Dubai business: scope before you buy

An AI receptionist for a Dubai service business runs the intake loop: takes the first message, asks what you need, books into a real calendar, answers FAQ from your documented answers, routes with context. It hands off judgement calls, complaints, negotiation, and anything medical, legal or visa-related. Most Dubai deployments are text-first on WhatsApp, and that changes the cost

Below is the scope we run before anyone builds one of these. It separates what to automate, what to augment, and what to keep human, and it explains the two cost paths: text-first and telephony voice

What can an AI receptionist actually handle?

An AI receptionist handles the four tasks that make up most first touchpoints: intake, scheduling, FAQ, and routing. Intake means capturing who you are, what you need, and how to reach you. Scheduling writes into a real calendar, checks availability against real slots, and sends the confirmation. FAQ answers from a bank of reviewed answers you own, never improvising numbers or policies. Routing sends the conversation to the right person with the context attached

Everything in that list works on text channels today. On WhatsApp, a caller can type a request at 2 am and get a useful reply in seconds, and the reply uses the official Business Platform, which handles opt-in and templates (Meta: getting opt-in). The same intake flow appears in our WhatsApp lead qualification article, where portal enquiries become a qualified lead in under thirty seconds.

On our a property-management platform property build, the AI handled about 60% of tenant chats without a person stepping in, on an MVP that launched in 10 days. The remaining 40% went to people, which is the part most demos leave out. A receptionist answers the first ring; it does not finish every conversation

What should hand off to a person?

Anything that needs judgement, a licence, or a signature hands off. Complaints go to a person. Price negotiation goes to a person. Anything medical or legal goes to a person. Visa and Emirates ID questions go to a person, because a wrong answer costs more than a slow human. So does a caller who asks, twice, to speak to someone

The rule we run is: the receptionist books the call, it does not negotiate. It confirms what it knows and passes the thread so the caller never repeats themselves. The agent opens with "I see you asked about the Marina Gate unit and the February viewing slot." That continuity is what makes the automation invisible, and it is the mechanism we described in human handoff on WhatsApp

A receptionist also must not invent answers. If the call is outside its documented FAQ, it says so and routes. A model that guesses a price or a refund policy does more damage than a slow human, and the damage is worse in Arabic, where the caller cannot tell how confident the answer is

The scope checklist: automate, augment, keep human

This is the artifact we fill in before pricing a receptionist build. It forces the decision per task, and it is where most of the value and most of the disagreement live. Print it, score every task, and you have the build scope and the cost driver in one table

Task	Automate	Augment	Keep human
First reply to a new WhatsApp/call	Yes, instantly	Reply with the caller's name and reason
Capture name, need, contact	Yes	Confirm parsed fields back to the caller
Booking a slot into the real calendar	Yes, against live availability	Propose 3 slots, write on choice
Rescheduling and cancellations	Yes, with window rules		Outside the window, or duplicate bookings
FAQ from documented answers	Yes		Anything not in the reviewed bank
Route to the right department/agent	Yes, with context packet	Score by urgency	Named-human assignment if no owner
Complaints and escalations		Detect the tone, summarise	The person who can fix it
Price, payment, quotes		Pass the lead	Negotiation and final numbers
Medical, legal, visa, Emirates ID		Read the category, stop	A licensed or trained person
Arabic voice note (Khaleeji, street noise)	Only with tested dialect data	Transcribe, then confirm in text	Anything below confidence threshold

Augment means the AI gathers and structures, and a human decides. Keep human means the AI does not decide at all. The rows that fall in keep human set the handoff design, and the handoff design is what the build actually costs

We priced a scope exactly like this in our AI automation cost article: a WhatsApp-first intake-and-scheduling build is low five figures AED and ships in 3 to 6 weeks. It stays low because a text flow has no telephony engineering. The moment you add a voice layer, the price moves

Does an AI receptionist work in Arabic and English?

On text, yes, both languages, often in one conversation. A Dubai caller writes a mix the way they talk, "abi 2 bedroom fi JVC", and a bilingual model with a reviewed Arabic and English prompt set handles it without forcing the caller to pick a language. This is the same language handling we describe in our Arabic-English voice agent article, and text is the part that is proven

On voice, it is harder, and honesty matters here. Standard Arabic and spoken Gulf Arabic are different. Khaleeji phrasing, dialect words, and street noise make transcription unreliable, so a voice receptionist needs dialect test data and a confidence threshold before it goes live. Below the threshold it confirms in text or routes to a person. A voice receptionist that sounds fluent and then gets a city or a price wrong is worse than a receptionist that never claims to be fluent

For Dubai, the practical answer is: text-first in both languages now, and voice added only once you have an evaluation set from your real callers, because the model that handles one clinic or one brokerage in MSA can still misread the way your customers actually talk

How much does an AI receptionist cost in Dubai?

The cost splits on the channel. A text-first intake-and-scheduling build on WhatsApp is typically low five figures AED, so 15,000 to 40,000 AED by our AI automation cost table, plus running costs of 500 to 5,000 AED a month and a maintenance retainer. The running costs cover model API spend, WhatsApp template messages and hosting

A telephony voice stack costs more, and the model is not the expensive part. The extra money goes to telephony minutes, a voice provider and the streaming integration, the latency engineering that keeps the caller from hearing a two-second pause, and the audio test set in the dialect you actually take calls in. The voice build also inherits the same PDPL data obligations on call recordings that apply to any personal data you hold under Federal Decree-Law No. 45 of 2021, which our PDPL-compliant AI CRM guide walks through.

If most of your calls arrive as WhatsApp messages anyway, the text build answers them faster and costs a fraction. If your business genuinely survives on inbound phone calls in a specific dialect, the voice build is the real product and should be priced as one, with the test set in the quote

On the a property-management platform property build the AI handled about 60% of tenant chats without a person stepping in, and the MVP shipped in 10 days. We quote the 40% that still needs a human, because that is the part that decides whether the handoff is clean or the caller repeats themselves

Will customers know they are talking to AI?

They should, from the first line. A receptionist that opens with "Hi, I am the automated assistant for Al Amal Medical Centre, how can I help?" sets the right expectation, and Meta holds WhatsApp businesses to a quality standard where pretending to be human is a fast way to damage it (Meta: Business Messaging Policy). The disclosure also protects you legally, because a caller who thought they were talking to a person can dispute whatever the bot said.

Disclosure does not cost you leads. Dubai callers already deal with automated phone trees and chatbots every day. What they punish is a bot that plays human and then has no answer, because now they have lost trust and their time. State that it is AI, be useful in the first two messages, and hand off fast when it stops being useful. The receptionist's job is to make the human takeover feel like the call is getting better, not getting transferred

The takeaway

Scope before you buy. Automate intake, scheduling from reviewed answers and FAQ; augment routing and tone detection; keep humans on complaints, prices, and anything medical, legal or visa-related. Run it text-first on WhatsApp in both languages before you pay for telephony voice, and disclose the AI in the first line. The build is low five figures AED when the handoff design is decided up front, and the cost grows with every row you try to push into automate before the staff and the test data are ready

Book the free audit and we will take your real first-touch conversations, fill in the automate, augment and keep human table, and give you a fixed number for the first build