Everyone here has already run a pilot
Ask around Dubai, Abu Dhabi or Sharjah and almost every company of any size will tell you it has tried something with AI. A chatbot on the website. A summarisation tool someone in marketing pays for monthly. A proof of concept a vendor built and demoed once. Far fewer can tell you what is running in production, who owns it, and what it changed.
That gap is the work. Not the model and not the tooling — the distance between a demo that impressed a room and a process that runs every day without anyone thinking about it.
How I work
1. Find the process worth automating
Not the most exciting one. The one with volume, clear rules, a measurable cost and an owner who wants it fixed. It is usually unglamorous: quote preparation, document intake, first-line support, reconciliation, tender paperwork.
2. Prove it on your data, not a demo set
A model that scores well on a vendor’s sample still has to meet your abbreviations, your bilingual documents, your scanned PDFs and your exceptions. We test against those before anyone signs anything.
3. Ship it where the work already happens
Inside the CRM, the inbox, the WhatsApp thread, the ERP — not in a separate portal your team has to remember to open. Adoption fails on that detail more often than on model quality.
4. Hand it over
Someone on your side has to be able to change a prompt, read the logs and switch it off. If the only person who understands it is me, I have built you a dependency rather than a capability.
What this usually looks like
- An assistant that answers customer questions from your own documents, in Arabic and English, and escalates cleanly when it should not answer
- Document intake — invoices, contracts, delivery notes — read, checked and filed without manual re-keying
- Internal search that finds things across years of shared drives
- Drafting tools for teams who write the same categories of document over and over
- Agents that take an action in another system, with an audit trail and a stop button
Arabic is not an afterthought
Most tooling is built and benchmarked in English. Arabic brings dialect, transliterated names, mixed-script documents and right-to-left rendering that breaks interfaces built without it. If your customers write to you in Arabic, the system has to be tested in Arabic before it goes live rather than translated afterwards.
Data, and the PDPL
Before anything is connected: what data leaves your systems, where it is processed, how long it is kept, and what your obligations are under the UAE’s Personal Data Protection Law. Some workloads should not go near a public API at all, and that belongs at the start of a project rather than in a security review three weeks before launch.
How we start
A short paid discovery, usually a week or two, ending in a written recommendation: the one or two processes worth automating first, what each would cost to build and to run, what it would save, and what could go wrong. If the honest answer is that AI is not your bottleneck, that is what the document will say.
Related reading
Most UAE AI pilots never reach production sets out the four places projects stop, and what the UAE PDPL asks of you before you connect a model is the data checklist I work through before anything is wired up. If the project is an assistant — for customers, for staff, or on WhatsApp — the conversational AI consulting page sets out how that work runs. And where the decision is settled and the answer is to build, the specification and delivery are described under AI development.