An AI agency in Morocco should be chosen on verifiable evidence, not on a demo. Since every digital vendor added "AI" to its deck, the word no longer separates anyone, and a business owner comparing three proposals reads the same vocabulary three times. This article sets out the checks a Moroccan business owner can make personally, without being technical, before committing a budget to an AI agency. Each one is answered by a link, a document or a name. None of them is answered by a promise.
What does an AI agency actually do?
An AI agency is a team that designs, builds and maintains software systems in which part of the work is done by an artificial intelligence model. That work takes three shapes: assistants and agents that answer, qualify or route instead of a person; document processing, where the system reads, extracts, classifies and retrieves; and the automation of repetitive decisions, where the system sorts, prioritises or shortlists. An agency that writes social posts with ChatGPT provides a real service, but it is a different trade: an AI agency delivers a system that runs in your business, connects to your existing tools, and stays someone's responsibility when it gets something wrong at three in the morning.
Why is choosing an AI agency in Morocco so difficult right now?
The difficulty is that no visible signal settles the question before you verify it yourself. Calling yourself an AI agency is a matter of vocabulary, and the usual trust signals — a polished website, a wall of logos, a partner badge — now cost almost nothing to produce. They no longer tell you who ships and who sells.
Google rankings do not help either. On 6 November 2026, the query "agence ia maroc" typed from Morocco in French opened on a Google AI Overview, followed among other results by articles ranking agencies against each other. Ranking well on that query proves a vendor can write about SEO, not that it can ship a system into production. Measured 6 November 2026; a results page changes, so run the check yourself.
How do you verify that an AI agency has shipped something to production?
Ask the AI agency for a product in production that you can open yourself — a web address, an App Store or Google Play listing, a trial account — rather than a demo video. A demo runs on examples chosen by the person showing it; a production system faces real users, real typos, outages, load and six months of maintenance.
Four follow-up questions then separate the two: who uses it today, since when, who maintains it, and what happened the last time the underlying model changed version? The last one only has an answer if the agency lived through it.
Artcom Solution answers that check with its own software: Gestio, DEVOXA and Hospikit are not client commissions but the agency's own products, which it builds, runs and maintains itself, and which you can open without an appointment. That is the evidence to require from any vendor, this one included.
Who owns the code, the data and the accounts when the project ends?
The source code, the data and the hosting accounts of an AI project should belong to you when the project ends, and the contract should say so in those words. "You will have access" is not "you own it": access can be withdrawn.
Four points are checkable in a proposal. The code repository is opened in your company's name, or transferred to you at delivery. The hosting and model-access accounts are opened in your name and billed to you. The database is exportable in an open format without the agency's involvement. And the instructions, configurations and datasets that shape the system's behaviour are listed as deliverables, exactly like the code.
The question to ask as written: if we stop working together tomorrow, what do I take with me, in what format, and how long would another team need to pick it up?
What does Moroccan law say about the data in an AI project?
In Morocco, the processing of personal data is governed by Law No. 09-08, enacted by Dahir No. 1-09-15 of 18 February 2009, with the Commission Nationale de contrôle de la protection des Données à caractère Personnel (CNDP) as the supervisory authority. An AI project that reads customer records, messages, CVs or case files falls inside that scope from its first row of data.
Three questions therefore belong in the contract, before they belong in the architecture. Which personal data will the system process, and which of it can you avoid handing over at all? Where is that data stored, and is it sent to a third-party model provider — which one, under what contract, and may that provider reuse it to train its models? Finally, are the CNDP formalities planned, and who handles them? If your customers are established in the European Union, the GDPR applies on top of Moroccan law.
Sources: the text of Law 09-08 and the CNDP website (cndp.ma), consulted on 6 November 2026. This article is not legal advice; have a real project reviewed by counsel.
Which recurring costs should an AI agency disclose before you sign?
An AI project carries costs that do not stop at delivery, and a serious agency writes them into the proposal instead of letting you discover them on the first invoice. Five items appear in almost every project: model calls, billed by usage and therefore rising with the number of users and the length of each exchange; hosting and the database, which run at night too; maintenance, which becomes compulsory when the provider changes or retires a model version without asking you; human supervision, needed the moment the system gets something wrong and someone has to catch it; and the data itself — cleaning, formatting, labelling — which is paid in time, often your team's.
The request to make: an estimate of running costs at three levels of usage, and what happens if the highest one is exceeded.
How do you tell whether the proposed AI really solves your problem?
An honest AI agency starts by checking whether your problem needs AI at all, and says so when it does not. Many requests phrased as "we need an AI agent" are solved by a better-designed form, a business rule, or an integration between two tools you already pay for: cheaper, more reliable, easier to maintain.
AI earns its place when three conditions meet: the input is messy — free text, documents, mixed languages — the volume is beyond what a person can handle by hand, and an occasional mistake is recoverable. The test to run: ask the agency to describe the non-AI version of your project, then explain why it is not enough. An agency that cannot describe that version is selling a tool, not a solution.
How does an AI agency measure whether the project succeeded?
Success on an AI project is defined before development, as a number you can track yourself: a processing time, a share of requests handled without human intervention, a count of qualified files per week. If that number is not written into the proposal, nobody will be able to say in six months whether the project worked.
Ask as well how quality is tested, because an AI system is not tested like a website. A team that knows how will talk about a set of real examples with expected answers, a check replayed at every new model version, an acceptance threshold, and a procedure for when the system is wrong: who sees the error, how fast, and how it gets corrected.
Which languages does your AI system have to handle?
In Morocco, an AI system facing the public has to handle French, Arabic and usually Darija written in Latin characters, because that is how customers actually write. An assistant tested on well-formed sentences collapses on a real message: abbreviations, two languages in one sentence, no diacritics, a transcribed voice note.
So have the system tested on your own messages, taken as they are from your WhatsApp or your inbox, and check Arabic end to end: the right-to-left interface, but also invoices, PDFs and exports. An agency that has already delivered in Arabic knows these traps; the others discover them at your expense.
What should you ask on the first call?
Ask the eight questions below as written, and note which ones get no answer in writing — those are the ones that become expensive later.
- What have you put into production that I can open right now?
- Who uses it, and for how long?
- Who owns the code when the project ends?
- Which personal data does the system process, and where is it stored?
- Which costs continue after delivery?
- What would the non-AI version of this project look like?
- How will we know in three months that it worked?
- What happens when the system gets something wrong?
A serious proposal puts those answers in writing, together with the exact scope of the first version, what is excluded from it, a timeline with milestones, the named external dependencies and the success metric as a number.
Where should you start?
Start by identifying what your business should fix first, before choosing an AI agency. That is what Artcom Solution's Growth Blueprint is for: an audit of your online presence and a roadmap, free and delivered in 72 hours, which you keep whether or not you go further. If your project is a product to build, the matching door is Custom Tech & AI; if you are already online but stuck, it is Scale Up. The delivered projects read in the same spirit: what was built, and what is live today.