A WhatsApp bot for a law firm in Dubai does one job: intake. It asks what the matter is, which jurisdiction it sits in, and who is on the other side. It screens those names against your existing client and matter list for conflicts. It states your consultation fee and sends a payment link inside the same chat. Then it writes the confirmed slot into the lawyer's calendar. It never answers a legal question. Published fee guides from Dubai firms put an initial consultation at roughly AED 500 to 1,500, so every enquiry that sits unread from Friday evening to Sunday morning is a paid slot that did not get sold. Build time is about two weeks. Our price is open: from AED 6,000 setup plus AED 1,200 a month.
What a legal intake chatbot in the UAE actually handles
A legal intake chatbot in the UAE is an automated WhatsApp conversation that collects matter details, screens for conflicts of interest, quotes and collects the consultation fee, and books the appointment, without giving any legal opinion.
That definition is deliberately narrow. Everything a Dubai firm needs from automation happens before the lawyer opens their mouth, and none of it is regulated work.
The intake sequence we build for a firm runs like this. Someone messages the firm number at 22:40. The bot replies inside a minute, asks whether the matter is commercial, employment, family, real estate or criminal, and branches from there. For an employment matter it asks whether the contract is with a mainland company or a free zone entity, whether the person is still employed, and whether there is a written termination notice. For a commercial dispute it asks the contract value band, the governing law clause if the client knows it, and the counterparty's registered name.
Then it runs the conflict screen, quotes the fee, takes payment, and books. The whole thing takes the client four to six minutes on their phone at midnight. The alternative, which is what most firms in Dubai run today, is a website form that lands in a shared inbox, gets read on Sunday morning, and gets a callback on Sunday afternoon by which time the person has already paid someone else.
The boundary: intake is not legal advice, and the bot has to know that
This is the part I would not let a vendor hand-wave. In the UAE, advocacy and legal consultancy are licensed professions. Federal Law No. 34 of 2022 regulating the advocacy and legal consultancy profession came into force on 2 January 2023, and in Dubai the Legal Affairs Department maintains the Roll of registered advocates and legal consultants and licenses the firms. Practising legal consultants have to be registered on that Roll. A chatbot is not on any roll, and it is not going to be.
So the bot's script has a hard floor built into it. It captures facts and refuses interpretation. Some real examples of the refusal path we write:
- Client asks "do I have a case?" The bot answers that only a registered legal consultant at the firm can assess the merits, and offers the next consultation slot.
- Client asks "how much compensation will I get for arbitrary dismissal?" The bot does not name a figure. It logs the question verbatim so the lawyer sees it before the call.
- Client asks "is my non-compete enforceable?" Same path: the bot logs the question and offers the slot anyway.
- Client pastes a contract PDF. The bot files it against the matter record for the lawyer, confirms receipt, and says nothing about its contents.
The design rule we use is that the bot may state facts about the firm and about the process, and nothing about the law. Fee, availability, documents needed, office location, languages spoken, how long a labour case usually takes to file at the relevant court, yes. Whether a specific clause holds up, never. Every one of those refusals is a scripted branch with a fixed answer, not a language model improvising a hedge. That distinction matters: an improvised hedge is one bad generation away from being a legal opinion with your firm's name on it.
The same architecture sits behind our WhatsApp AI sales agent, where the equivalent rule is that the agent never confirms a payment it has not seen land. Different domain, same principle. You decide in advance what the machine is not allowed to say, and you enforce it in the flow rather than in the prompt.
What a slow intake costs a Dubai firm
Run the arithmetic on your own numbers rather than mine. Take a firm that gets 40 enquiries a month across the website form, Google Ads and the WhatsApp number on the Google Business Profile. Say a third of them arrive outside office hours, which is conservative in a city where people research a dispute at night. That is 13 enquiries a month whose first human contact is the next working morning at the earliest, and Friday evening enquiries wait until Monday.
At the published AED 500 to 1,500 range for a paid initial consultation, losing four of those thirteen to a faster competitor costs the firm AED 2,000 to 6,000 a month in consultation revenue alone, before you count the matters those consultations would have opened. A single commercial retainer that walks is worth more than the whole year of automation.
There is a second cost, and it is the one owners underestimate. Somebody at the firm is doing the intake by hand: re-typing name, nationality, matter type and counterparty into a spreadsheet, then into the case management system, then chasing a bank transfer for the consultation fee. In our own modelling of Dubai SMB back office work, that pattern runs around AED 2,000 a month of salaried time spent on work a machine should do, with roughly another AED 4,100 a month leaking through slow and missed leads. A law firm sits at the high end of that because the intake record has more fields than a salon booking.
Count your own after-hours enquiry share for one month. That single number decides whether any of this pays back.
Conflict-of-interest check before the calendar slot, not after
This is the feature that separates a real law firm client intake automation build in Dubai from a generic booking bot, and it is the one nobody outside the profession thinks to ask for.
A conflict discovered after you have taken a client's money and heard forty minutes of confidential background is an expensive, embarrassing problem. A conflict caught during intake is a polite decline. So in a legal build we move the check ahead of the calendar, not after it.
Mechanically it works like this. The bot asks for the full legal names of every party involved: the client, the counterparty, and any related entities. It normalises what it gets, because people type company names four different ways and Arabic names transliterate inconsistently. Then it queries the firm's own client and matter list, which usually lives in the case management system or the CRM, and returns one of three states.
Clean, and the bot proceeds straight to fee and booking. Possible match, and the bot stops booking, tells the client the firm needs to complete a standard check before confirming, and routes the record to the partner or the compliance contact with the matched names attached. Confirmed conflict, and the bot never books at all; a human sends the decline.
The bot does not decide. It flags and routes. That distinction keeps the professional judgement where it belongs and still removes the twenty minutes a junior would otherwise spend on the search. For a DIFC or ADGM practice with an institutional client base, that screen is not a nice-to-have. It is the reason the intake is allowed to be automated at all.
The consultation is paid, and the bot should say so
Most of the intake automation content written for law firms comes out of the United States, where the free consultation is the default and the whole funnel is designed to give something away. That model does not transfer cleanly here. Plenty of Dubai firms charge for the first meeting, and published fee guides on firm websites, including HHS Lawyers and Al Basti Advocates, quote around AED 500 to 1,500 for an initial consultation and higher for senior specialists or complex matters. Treat those as published guide figures rather than a market survey, and check what your own competitors advertise this quarter.
A paid first consultation is a gift to an automated intake, because it gives the bot a real qualifying gate. The sequence we run:
- Bot states the fee plainly once the matter type is known, in the same message as the available slots.
- Client accepts, and the bot sends a payment link generated against that specific consultation.
- Payment lands, the webhook fires, the slot is confirmed and blocked in the calendar.
- No payment inside the window, the slot releases automatically and the bot follows up once, then stops.
Nothing here is exotic. It is a payment gateway, a calendar and a webhook wired together properly, which is exactly what our AI automation work is. The interesting part is what it filters out. Someone who will not pay AED 750 to have a lawyer look at their matter was never going to sign a retainer, and every hour the firm does not spend on that person is an hour on a real file.
One caveat from experience: do not let the bot negotiate the fee. If a client pushes back, the bot holds the number and offers to have a partner call. A machine that discounts is worse than no machine.
Bilingual intake, and why translation is not enough
Dubai clients switch between Arabic and English inside a single message, and legal vocabulary is exactly where a translation layer falls over. "قضية عمالية" needs to route to employment, not to a generic "case" bucket. "شيك بدون رصيد" needs to route to the bounced-cheque flow with the right urgency, because that client has a deadline. A bot that machine-translates the Arabic into English and then classifies the English will misroute a meaningful share of those.
What works is writing both language paths natively, with the legal terms mapped by hand in each. That means an Arabic-speaking reviewer goes through the classification list once and tells you which phrases real clients use, which is a day of work that pays for itself in the first month. It also means the bot answers in the language the client opened with, and holds it, rather than drifting back to English three messages in because the model finds English easier.
Same for the formality register. Arabic intake for a family matter reads differently from Arabic intake for a corporate dispute, and getting that wrong costs you the client silently. Nobody complains. They just stop replying.
Compliance: WhatsApp Business Platform, opt-in and PDPL
Three constraints shape the build, and all three are cheap if you handle them at the start and expensive if you discover them in week two.
First, the channel. This runs on the WhatsApp Business Platform through an official provider, not the free WhatsApp Business app. Meta restricts what it calls Regulated Verticals, prohibits messaging about them on the Business app entirely, and permits them on the Platform only in limited countries and subject to applicable law. Legal services are not named in the public examples I could find, which is precisely why you get your provider to confirm your vertical in writing before you build. I would not take a forum post as the answer on this.
Second, message timing. Under Meta's rules, a business can reply freely for 24 hours from the client's last inbound message. Outside that window, the firm may only send an approved message template, and Meta can review, pause or reject any template at any time. For a law firm that shapes the follow-up design: the consultation reminder and the "your documents are still outstanding" nudge both need approved templates sitting ready, and template approval belongs at the start of the project timeline rather than the end.
Third, data protection. Intake data is personal data by any definition, and a matter description is sensitive by any reasonable reading. For a mainland Dubai firm the reference point is Federal Decree-Law No. 45 of 2021 on the Protection of Personal Data, in effect since January 2022. Its consent standard is specific, informed and unambiguous, given by a statement or a clear affirmative action, withdrawable at any time. Before processing, the controller has to tell the data subject the purpose, who the data will be shared with, and the safeguards around any cross-border transfer. It also requires technical and organisational security measures proportionate to the risk.
Firms inside the DIFC sit under a different instrument: DIFC Data Protection Law No. 5 of 2020, enforceable since October 2020 and amended in July 2025 with the amendments taking effect on 15 July 2025, which broadened its territorial reach to processing carried out in the DIFC regardless of where the controller is incorporated. ADGM runs its own regime as well. You know this better than I do; the reason it appears in a build document is that it changes where the intake record is allowed to live and who may reach it.
In practice, the design decisions that follow are simple. The opt-in is an explicit line in the first bot message, not a buried checkbox. The intake record moves out of the conversation layer into the firm's own case management system or CRM as soon as it is complete, and the bot's working memory of it is cleared on the same trigger. Access to the record is scoped to the people who need it. Nothing sensitive gets written into a marketing tool.
Mainland, DIFC and ADGM firms need different intake trees
A common-law practice in the DIFC or ADGM and a mainland firm working under UAE civil law are not selling the same service, and their intake questions diverge from the second question onward.
For a DIFC or ADGM practice the early screen is jurisdictional. Which court or tribunal has the matter, is there an arbitration clause, is the counterparty a DIFC entity, does the contract opt into DIFC Courts jurisdiction. Get that wrong and the consultation is spent unpicking the confusion instead of on the matter.
For a mainland firm the early screen is procedural. Which emirate, is there a Dubai Courts case number already open, has a police complaint been filed, is there a deadline running. And the document list differs: mainland matters usually need the Emirates ID, the trade licence and the tenancy or employment contract, and the bot should ask for them by name and collect them in the chat rather than emailing a PDF checklist nobody opens.
We build these as separate branches, not one flow with an if-statement. It costs a couple of extra days and it removes the most common intake failure I see, which is a lawyer arriving at a paid consultation without the one document that would have made the call useful.
How we run this on ourselves
I would rather show the pipe than describe it. Our own WhatsApp agent handles the enquiry flow for the SkyLight studio business end to end: it answers in under a minute, qualifies, quotes, sends the payment link, and books, with no human in the thread. That contour carries a business doing about AED 67.6k net a month at roughly 5x ROAS on ad spend. It has closed deals solo, first message to paid.
The part law firm owners tend to care about most is not the speed. It is that every conversation is scored automatically for response time, whether the intake was completed properly, and whether the follow-up actually went out. Across a team, that is the difference between assuming your intake is being handled and knowing it. The machine has no off days. It does not decide an enquiry looked low value, and it does not forget the follow-up because it was busy. The reply at 03:00 on Saturday is the same reply as Monday at 10:00.
Our pricing sits in the open on the pricing page: AI automation from AED 6,000 setup plus AED 1,200 a month for one process running unattended, wired into whatever you already have. If you want the leak measured before you commit to anything, the Growth Audit starts at AED 3,000 and produces the numbers for your own firm: after-hours enquiry share, response times, and where the intake breaks.
