AI agency in Morocco: the questions to ask before you sign
Before handing an AI project to an agency, ten questions on quality measurement, data, code ownership and payment. And what a reasonable first engagement costs in DH.
A reasonable first engagement with an AI agency costs between 15,000 and 30,000 DH excl. VAT for an audit, or 12,000 to 60,000 DH excl. VAT for a prototype on your documents: that is the price list we publish, and it is the order of magnitude to compare against. Before signing for more, ask the ten questions below. A serious agency answers them in writing; an agency that dodges them is already telling you what you need to know.
We are ourselves a software development agency in Casablanca that does AI. These questions therefore apply to us too, and we give our own answers.
Updated on 17 November 2026.
1. On which documents will you measure quality?
The most important question. A success rate measured on demo documents tells you nothing about your invoices, contracts or forms. Insist that quality is measured on a sample of your own documents, field by field, before you fund the integration. Our answer: that is exactly what the prototype is for, and if the quality is not there, you know before paying for the rest.
2. What error rate do you announce, and where does it come from?
Be wary of a round figure given before anyone has seen your documents. AI makes mistakes; the rate depends on scan quality, the variety of layouts and the language. An honest agency refuses to promise a universal percentage and offers to measure it. An agency promising “99%” without a sample is selling a sales pitch.
3. What happens when the AI is not sure?
Ask to see the validation screen. A usable process shows the document next to what the AI read, highlights the doubts, and lets a person correct them. Without it, errors enter your software without anyone seeing them. All our integrations keep human validation on what matters.
4. Where does my data go, and who keeps it?
Many projects rely on a model hosted abroad. Sending personal data to such a model is a transfer governed by Moroccan law 09-08: the agency must be able to say so, offer to mask or minimise the data, and help you prepare the technical part of the filing. The details are in our guide AI and law 09-08. Also ask whether your data is used to train a model, the agency’s or the provider’s.
5. Will you “train a model” for me?
If the answer is yes, ask on what data, how many examples, who labels them, how long it takes and how quality will be compared with an existing model without training. For most business projects, an existing model used well is enough, and that is what we do: we do not promise custom model training. A promise of “proprietary AI” on an SME budget deserves questions.
6. Who owns the code and the data?
Have the answer written into the contract. With us, custom development belongs to the client once paid: code, database and access are handed over. Our own products (Médiviz, Residapp) remain our property and are licensed, but the client’s data stays with the client. Be wary of a platform from which you can never export your data or recover the rules and instructions given to the model.
7. How does the AI connect to my software?
AI isolated in a separate tool creates double data entry. The right question: how does the result reach your ERP, CRM or back office, through which API, with which permissions? See AI in your existing software. If your software has neither an API nor an accessible database, an honest agency will tell you before quoting.
8. How much will model usage cost each month?
On top of development, an external model is paid by volume. Ask for an estimate on your real volume, measured on the sample, and a usage cap. For high-volume processing, it is often this cost, not development, that decides profitability.
9. Which projects can you show, and which are AI?
Ask for verifiable projects and have the agency spell out what in each one is really AI. Our answer: our published case studies, such as Doctifile (patient journey and flow of medical documents) or LA MAS (health cover platform for the insurer side), are business platforms, not AI projects. They show that we can build and run the software AI fits into; they do not prove AI performance, and we do not present them that way. Our AI projects are our own products, described with their real figures and their limits: ScanClear (website audit with AI hosted on our servers), Isnad (reading public tender files), MeetScribe (meeting minutes in Darija) and PulmoDeep (prototype to assist reading chest X-rays).
10. How is payment structured, and what happens after go-live?
A payment schedule tied to deliverables protects both sides. Ours: 40% at signature, 30% at the first usable version, 30% at go-live, then three months of defect fixes included. Also ask who monitors quality over time: your documents change, a supplier changes its invoice, and reading quality can drop without anyone noticing.
What budget should you plan, step by step?
| Step | Range (DH excl. VAT) | What you get |
|---|---|---|
| AI audit | 15,000 – 30,000 | Realistic use cases, the data involved, where the model runs |
| Prototype | 12,000 – 60,000 | One use case tested on your documents, with quality measured |
| Integration into a workflow | 18,000 – 200,000 | AI connected to your software, human validation and logging included |
| Dedicated team (if needed) | from 2,500 / day / profile | Developers assigned to your project over time |
Source: our price list. Assumptions: one process at a time, model usage fees extra, billed by volume by the provider. No quote above 100,000 DH without prior scoping.
When should you not sign with an AI agency?
When the need is covered by off-the-shelf software: a SaaS invoice-reading tool built into your accounting can be enough for medium volume and standard suppliers. When the volume is low. When the process itself is not defined. In those cases, an agency pushing you towards a custom project is defending its revenue, not your interest. The article AI or automation helps you sort it out.
To see the four use cases we handle: document reading, checks and validation, bulk processing, AI in your software. For a first conversation: request a quote.
Frequently asked questions
Do you need a specialist AI agency or a development agency?
In a business project, AI is one step in a software process. A team that can build the whole process, integration and validation included, is generally more useful than a specialist in the model alone.
Can you ask for a free prototype?
A serious prototype measures quality on your documents, which takes time. When it is free, it is usually done on demo documents and proves nothing.
How long does a prototype take?
It depends on the use case and on when the sample is available; the timeline is set during the audit.
Related articles
Artificial intelligence
How much does an AI project cost in Morocco?
From 12,000 DH excl. VAT for a prototype to 200,000 DH for AI built into a business workflow. The four cost drivers: volume, model usage, integration, human review.
Artificial intelligence
AI or automation: how to tell what you are being sold
Much of what is sold as “AI” is rule-based automation, and that is not a flaw. A simple test to tell the difference, and why it changes the price and the risk.
Artificial intelligence
Automated invoice and document reading: what works, what fails
Automated reading works well on clean digital documents, much less on damaged scans or handwriting. Type by type: what works, what fails, and where a person must check.
Newsletter
Get our upcoming articles straight to your inbox.
