Artificial intelligence
Some links on this page are affiliate links: we may earn a commission at no extra cost to you. Our approach
AI is not an isolated technical project: it is a way of automating certain knowledge tasks. For a company, the useful question is not 'which AI should we choose' but 'which concrete problems will we measure before and after'. A customer request handled twice as fast, document sorting automated, a slightly more accurate stock forecast: that is what a successful AI project looks like. ## Three families of use Generative assistants help to write, summarise and analyse. Classification and prediction models, from machine learning, are used to sort, anticipate and score. AI agents, more recent, chain several actions with tools to complete a whole task. Each family requires a different level of data and control. ## The real cost: integration and verification An AI tool costs little in subscription, a lot in integration and review. Allowing human time to check results, plus a policy stating what is never put into a tool, is often more profitable than a more powerful model. ## Responsible and useful AI Sensitive decisions (accepting a customer, refusing credit, scoring a supplier) must remain explainable and verifiable. For everything else, the right measure is simple: time saved without loss of quality, and teams that know when the AI is wrong. ## A concrete case: customer service A distribution company receives hundreds of messages a week. An AI assistant classifies requests, drafts a reply in French and Arabic in the brand's tone, and forwards it to the adviser. The adviser re-reads, corrects if needed and sends. The gain is measured in minutes per message and response time, provided a threshold is set above which nothing is sent without review.When this guide is for you
- Your teams spend several hours a week writing repetitive content (customer replies, product descriptions, summaries)
- You receive a lot of messages and documents and want to classify them or extract the essentials automatically
- You want to test AI on concrete cases before committing to a large investment
- You hear about generative AI everywhere and need clarity, without spending your nights on it
- You need to set rules for how your staff use AI tools to avoid data leaks and errors
Step by step
- Identifying use cases — List the company's repetitive, costly or slow tasks, then assess each one's potential gain and risk. Choose two or three concrete use cases rather than launching an abstract 'AI' programme.
- Choosing tools and models — Compare models and tools (conversational assistants, model APIs, cloud platforms) on cost, language, confidentiality and existing integrations. A general-purpose tool is often enough to start.
- Prototype and evaluation — Build a minimal version of one use case, test it on real data and measure the gain: time saved, quality produced, error rate, number of human corrections needed.
- Industrialisation and integration — Integrate the use into the company's tools (CRM, helpdesk, e-commerce), plan for human validation and monitor quality and costs. A prototype that stays outside the daily tools is useless.
- Governance, training and monitoring — Write a usage policy (forbidden data, checking results, responsibility), train the teams and review results every month. AI amplifies good and bad practices alike.
What it costs (in MAD)
| Service | Indicative price | Good to know |
|---|---|---|
| Scoping and use-case workshops | 8,000 – 30,000 MAD (≈ 720 – 2,700 €) | One to three working days with the business teams |
| Generative AI prototype | 15,000 – 60,000 MAD (≈ 1,350 – 5,400 €) | One use case, tested on real data |
| AI assistant subscription (ChatGPT, Claude) | 150 – 250 MAD per user per month (≈ 14 – 22 €) | Business plans, billed in foreign currency |
| Model API (usage) | 500 – 10,000 MAD per month (≈ 45 – 900 €) | Billed by volume of tokens or requests |
| Integration into an existing tool | 30,000 – 150,000 MAD (≈ 2,700 – 13,500 €) | Depends on the complexity of existing systems |
| Team training | 5,000 – 25,000 MAD (≈ 450 – 2,250 €) | Per session or as a programme over several months |
| Support and governance | 10,000 – 40,000 MAD per month (≈ 900 – 3,600 €) | Usage policy, monitoring of use cases and costs |
Indicative ranges for the Moroccan market in 2026, excluding tax. Prices vary with scope and provider.
Mistakes to avoid
- Starting with the technology instead of the business problem, and ending up with a prototype nobody uses
- Pasting confidential data (customers, contracts, HR) into consumer tools without checking the terms of use
- Trusting answers blindly: a model can invent plausible but false information
- Measuring success by the number of people trained rather than by time saved or quality produced
- Neglecting the legal framework and responsibility for decisions that affect customers
Frequently asked questions
Where should a small company start with AI?
With one specific task that wastes time every week: answering frequent questions, writing product descriptions, summarising complaints. First train a few people on these tasks, measure the gain, then expand.
Will AI replace employees?
It mainly changes the nature of the work: repetitive actions shrink, while checking, customer relationships and judgement take up more space. Companies that benefit most are those that train their teams to verify and enrich AI output.
Can AI be used in Morocco without particular constraints?
Law 09-08 on the protection of personal data applies as soon as AI processes data on individuals, whatever the processing or hosting. A processing declaration and informing individuals are generally required, and some sectors are more tightly regulated.
What is the difference between generative AI and machine learning?
Generative AI produces text, images or code from prompts; machine learning learns models from historical data to predict or classify. The two can be combined, but the skills and data required differ.
Does AI improve search rankings?
Indirectly. Clear, structured and reliable content is more likely to be cited by AI engines. AI can help produce that content, but without human review and real expertise it produces generic text that engines and readers spot quickly.
Updated on
Related guides
Version française: /services/data-ia/intelligence-artificielle/ · AR · NL