Loading
Novamind scopes, builds and runs conversational AI for businesses in Kuwait, the UAE, Oman, Saudi Arabia, Qatar and Bahrain: chatbots that answer customers on WhatsApp, on your website and inside your app, in the Arabic your customers actually write and in English. The team works remotely with clients across the GCC. There is no office to visit and no field staff; discovery, build, testing and support happen over calls, a shared repository and a staging bot you can message at any time.
Owners and operations leads who handle customer questions on WhatsApp, on a website or in an app, and want a bot that answers correctly in the customer's language and hands over to a person when it should.
Examples of the problems this suits: a clinic in Kuwait City answering the same booking questions every day, a distributor in Sharjah fielding order status and stock questions, a facilities company in Muscat logging service requests, a training academy handling enrollment questions in two languages. In every case the bot is a front door to the systems and people you already have, not a replacement for them.
This page is written around Kuwait, the UAE and Oman. The same approach applies in Saudi Arabia, Qatar and Bahrain, and the data handling section below covers each country's law. For the wider picture on what these systems can and cannot do, read our note on AI chatbots in customer service.
Five things, each scoped on its own so you pay for the channel and the integrations you need. Voice and IVR are not part of this service.
A bot on the WhatsApp Business Platform that answers questions, takes bookings or orders, checks status against your systems, and hands over to a person inside the same conversation. We prepare the message templates you need outside the service window in both Arabic and English, and we walk through Meta's business verification with you. We work directly against the Cloud API or through a provider you already use. Novamind is not a Meta Business Partner or Business Solution Provider.
A chat window on your site in Arabic (right to left) and English, with mixed direction text rendered correctly, that draws answers from your documentation, product data or an internal knowledge base. Built as a retrieval augmented generation (RAG) assistant, so answers come from your content rather than from the model's memory, with a link to the source page where that helps the customer.
The same assistant wired into your existing web or mobile app through an API, so it can act for a signed in user: show their order, book a slot, open a ticket. Read only access first. Write access only for the actions you have agreed, with every write logged.
Every bot we ship has a path to a person. Handoff can land in the inbox tool you already use, in a shared WhatsApp inbox, or in a simple queue we build. The bot passes on the transcript, the language the customer wrote in and what it already tried, so your agent does not start from zero.
CRM, ERP, booking, order and ticketing systems, including Axion ERP for clients who run it. We define exactly which records the bot may read and which it may write, and the difference matters: a read only bot can leak but cannot corrupt, a bot with write access needs guard rails and an audit trail.
Do not expect customers in Kuwait to write Modern Standard Arabic to a business. The bot has to read what they do write.
They write Kuwaiti or Gulf dialect, often with English product names in the same sentence, sometimes in Arabizi (Arabic typed in Latin letters and digits), and sometimes they switch language halfway through a conversation. A bot tuned only on formal Arabic, or on English that has been machine translated, reads these messages badly and answers stiffly.
Our approach is to build the test set from your real messages before choosing anything else. During discovery we export a sample of actual conversations or tickets, group them into intents, and record how each was written: dialect, Modern Standard Arabic, English, Arabizi or a mix. That set becomes the acceptance test for the bot, so the question of whether it understands your customers is answered by a count of passed and failed cases you can inspect, not by a promise.
Two rules apply by default. The bot replies in the language and register the customer used: dialect gets a plain, polite dialect reply, formal Arabic gets formal Arabic, English gets English. And no dialect is described as fully supported, by us or on behalf of any model vendor. You get the evaluation results for your own message set instead, refreshed whenever the bot or the model changes.
On web and in app, Arabic also means right to left layout, correct rendering of mixed direction text (an Arabic sentence with an English SKU inside it), the numeral style your customers expect, and a widget that behaves the same whether the page is in Arabic or English.
For most GCC businesses WhatsApp is where customers already are, so the running cost of the bot is mostly Meta's messaging fees. This is how they work, from Meta's own documentation, checked on September 4, 2026.
Meta bills you directly for messages. Novamind does not resell messaging and does not mark it up; the fees appear in your cost model as dated figures from Meta's rate card so the monthly running cost is visible before you commit.
The same four phases as every Novamind engagement on the services page, applied to a chatbot.
Written brief and fixed price quote
A structured session, typically two to four hours, plus an export of your real conversations or tickets. We list the channels in use, the intents the bot must handle, which Arabic your customers write, the systems the bot must read or write, what data it may store and where, and the success criteria as measurable checks. Both sides sign the brief off before design begins. The discovery session is free.
Flows, templates and API contract
Conversation flows for each intent in Arabic and English, the handoff rules, the WhatsApp template set, and an API contract for every integration. Where there is a widget or an in app screen, high fidelity Figma screens for it, right to left included.
Staged bot every sprint
Two week sprints with a staging bot you can message from the first sprint. The evaluation set from discovery runs on every change, you have read access to the repository from day one, and a weekly async video update tells you what moved.
Production deploy and runbook
Production deployment in the agreed hosting region, error monitoring, alerts, a documented runbook, and a 30 day warranty covering bug fixes at no extra charge. After that, optional ongoing support on a monthly retainer.
Most clients start with a pilot: one channel, one language pair, a fixed intent list, a real test set, a human handoff path, transcripts stored in the agreed region, and a written report of what the bot answered correctly and what it escalated. A pilot is priced and scoped as a fixed scope project, and the report is the basis for deciding whether to extend.
AI and automation engagements start from AED 8,000, the published figure on our services page. That is the starting point for a straightforward build. Your fixed price is quoted in writing after discovery.
Free discovery session, then a fixed price in writing. No time and materials billing on a fixed scope build. A bilateral NDA is available before discovery.
Get a quoteRunning costs are separate from the build and are paid by you to the providers: Meta's per message fees on WhatsApp, model usage billed by the model vendor, and hosting. The quote lists dated figures for each so the monthly cost is on the table before you sign.
The same three models as every Novamind engagement.
What you keep: on final payment the code, prompts, templates, evaluation set and conversation data are yours, with no licensing hooks and no dependency on Novamind to keep the bot running. You can take the whole thing to another team at any point.
A chatbot stores personal data: names, phone numbers, transcripts, sometimes order or health details. Where it may be stored and how it may move is set by law in each country. The notes below summarize the sources listed at the end of this page, checked on September 4, 2026. This is not legal advice.
From the providers' own region lists: AWS runs Middle East (UAE), me-central-1, and Middle East (Bahrain), me-south-1, both opt in regions (AWS regions). Azure runs UAE North in Dubai, UAE Central in Abu Dhabi with restricted access, and Qatar Central in Doha (Azure regions). Google Cloud runs me-central1 in Doha and me-central2 in Dammam and lists no UAE region (Google Cloud locations); Dammam is open only to Saudi based customers buying through CNTXT or to accounts on invoiced billing (Google Cloud, Dammam access). None of the three lists a region in Kuwait or Oman, so a client there chooses the nearest permitted region and records that choice in the transfer disclosure. The region, the retention period and who can read transcripts are written into the brief during discovery.
The pieces a conversational AI build usually touches, drawn from the stack listed on the services page.
Model choice is made per project on the evaluation set, not by preference. Model usage is billed to you by the vendor at the vendor's published rates, which we read on the day of the quote.
Stated up front, so the brief starts from the right place.
We have not published a chatbot case study yet, and we would rather say so than invent one.
What we can show is a production system built and operated for a Gulf client: GSG Academy, the learning platform we built for German Standard Group, with an iOS and Android app, a web app, an admin panel and a NestJS back end, live at gsgacademy.co. It is a learning platform, not a chatbot project. It is here as evidence that the studio ships and runs systems that real users depend on, for a KHDA approved academy. Our two products, Ignis and Axion ERP, are the other evidence.
The questions we hear before a conversational AI project starts.
Send the channel, the language mix and the systems involved. We read every message and respond within 24 hours.
Ask a questionTell us which channel your customers use, which languages they write in, and which systems the bot would need to reach. We reply within 24 hours with a clear next step, usually a discovery session.