What does an AI agency actually do for a Moroccan business?+
It designs and deploys AI around how your business already works: multilingual chatbots (French, Arabic, Darija), AI agents that run a workflow end to end, process automation, RAG knowledge bases built on your own documents, and computer vision systems. For a Moroccan SME the usual result is repetitive work down by 40 to 60 percent, customer service that answers outside office hours, and decisions taken on current data instead of last month's report.
What does an AI project cost in Morocco?+
A WhatsApp or web chatbot runs 25,000 to 60,000 MAD depending on scope. End-to-end process automation, 40,000 to 120,000 MAD. A business AI agent with RAG over your documents, 80,000 to 250,000 MAD. On top of the build, budget 2,000 to 8,000 MAD a month in model costs (OpenAI, Anthropic) according to volume.
Which Moroccan cities do you work in?+
Our teams are based in Tangier and we work across the country: Casablanca, Rabat, Marrakech, Fez, Agadir, Meknes, Oujda, Tetouan, Kenitra, El Jadida and Nador. Projects run hybrid. Most need two or three visits in person (scoping, delivery, training) and the rest is remote, with a weekly check-in.
Does my data stay confidential?+
Yes, and it is not up for negotiation. For sensitive data we configure the APIs in zero retention mode (OpenAI Enterprise, Anthropic Workspaces, Azure OpenAI), where nothing you send is used to train a model. Where the constraints are tighter still, in health, defence or regulated finance, we deploy open-source models on private infrastructure. Our deployments comply with Moroccan law 09-08 (CNDP).
Which AI use case should a Moroccan SME start with?+
Three pay back fastest. A WhatsApp chatbot for lead qualification and first-line support: live in two to four weeks, returns visible inside three months. Automating accounting or administrative data entry with OCR and workflows. An internal AI agent for document search, with RAG over your contracts, quotes and procedures. What to avoid at the start is a custom ML project: those need clean datasets you probably do not have yet, and the payback is a long way out.