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.
An AI project applied to your documents costs between 12,000 and 60,000 DH excl. VAT for a prototype, and between 18,000 and 200,000 DH for integration into a business workflow in production, preceded if needed by an audit at 15,000-30,000 DH. On top of this build cost comes a model usage cost, which depends on your volume and is measured during the prototype.
Updated on 17 November 2026.
What kind of AI project are we talking about?
Not a generic assistant, nor a model trained for you. We mean AI applied to a precise task: reading documents (invoices, contracts, forms, supporting documents) and extracting their fields, checking a file before approval, processing volumes a team can no longer keep up with, or adding these functions to your existing software. We use existing models, available through an API or specialised; we do not promise to train a custom model.
How much does each stage cost?
| Stage | Range (DH excl. VAT) | Assumption | Source |
|---|---|---|---|
| AI audit | 15,000 to 30,000 | Inventory of documents and tasks, sample, recommendation: do it, don’t do it, or wait | Digi4 price grid |
| Prototype | 12,000 to 60,000 | One document type or one task, tested on a sample of your real documents | Digi4 price grid |
| Integration into a workflow | 18,000 to 200,000 | Connection to your software, review screens, log, error handling | Digi4 price grid |
| Model usage | Variable, by volume | Billed by the model provider according to the amount of text processed; measured during the prototype | Digi4 estimate (to be measured) |
| Human review | Your team’s time | A person reviews what the AI proposes, at least at the start; it is a real cost to count | Digi4 estimate |
The full price grid is on our pricing page. Payment follows the same rule as our other projects: 40% on signature, 30% at the first usable version, 30% at go-live.
What are the four factors that set the price?
1. Volume
Processing fifty documents a month and processing fifty thousand are not the same project. At low volume, the model usage cost is negligible and the main question is whether automation saves enough time to justify the build. At high volume, you need a processing queue, retries on errors, tracking of what went through and what failed, and the usage cost becomes a line in the annual budget.
2. Model usage
Models available through an API are billed by their provider on usage, according to the amount of text (and images) sent and produced. A two-page document read by a small model and a fifty-page file analysed by a large model do not have the same unit cost. That is why we give no price per document before the prototype: we measure it on your real documents, then choose the cheapest model that does the job properly. When data must not leave your infrastructure, a model hosted on your premises or with a hosting provider of your choice changes the equation: lower usage cost, higher hosting cost.
3. Integration
This is often the biggest item, and the least visible in demos. An AI that extracts fields from an invoice is useless if someone then has to retype them into the ERP. Integration covers collecting the documents (email, scan, portal), writing the results into your software, the screens where a person approves or corrects, and the log of what was done. If your software is recent and has an API, integration sits at the bottom of the range; if it is old or closed, it goes up.
4. Human review
An AI that reads documents sometimes gets things wrong. We do not publish a generic accuracy rate: it depends on your documents (scan quality, variety of formats, handwriting) and must be measured on a sample of your own during the prototype. What we know in advance is that you need review by a person, at least at the start, and rules to decide when the AI can go through without review. That review time is a cost; the real gain is the difference between today’s data-entry time and that review time.
When should you not do AI?
More often than people think. We advise against launching an AI project when:
- the volume is too low to pay back the build: a few dozen documents a month are quicker to type in than to automate;
- the documents are already structured (an export, a file, an API): a classic rule does the job, without AI, faster and cheaper;
- errors are unacceptable and no one can review;
- the process itself is unclear: automating a vague process just produces errors faster.
That is precisely the role of the audit: it can conclude that you should not do AI, and that is a useful conclusion that saves you a pointless prototype.
What should you prepare before asking for a quote?
Three things are enough for a serious first price range. A sample of real documents, anonymised if needed, that shows the variety of formats and not just the cleanest ones. The monthly volume, and its peaks. The software the results must land in, with what you know about its interface. With that, we can tell whether the project sits at the bottom or the top of the range, and whether it is better to start with an audit or go straight to a prototype.
What about personal data in the documents?
Invoices, medical records, identity documents: many documents contain personal data. In Morocco, its processing falls under Law 09-08, enforced by the CNDP (www.cndp.ma). Sending these documents to a model hosted abroad is a data transfer to be examined during scoping. This paragraph is information, not legal advice; our free CNDP check gives a first view.
What does AI look like inside business software?
In a case-management platform, like the one we built for the health insurer LA MAS (member, partner and back-office portals), the natural place for AI is in the document flow: read the attachment, pre-fill the file, flag what is missing, then let the case handler approve. AI is added to software that works; it does not replace it.
Frequently asked questions
Can we test without committing 200,000 DH?
Yes: that is the role of the prototype, from 12,000 DH, on a single document type. You see the result on your documents before deciding to integrate.
What success rate can you guarantee?
None before measuring. We measure on your sample during the prototype and give you the observed figure, with the failure cases.
Do you train a model for us?
No. We use existing models, with instructions, control rules and human review. It is faster, cheaper and easier to maintain.
Who pays for model usage?
You do, at the provider’s actual cost; the account can be opened in your name.
Service details: custom AI: documents, checks and processing. Describe your documents and your volume, answer within 24 hours.
Related articles
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
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.
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.
