AI consulting
Some links on this page are affiliate links: we may earn a commission at no extra cost to you. Our approach
AI consulting is not about choosing the most impressive model, but finding the tasks where an assistant brings a measurable gain without introducing new risks. The value comes from integration and verification, not from the technology itself. ## The AI that fails fastest Projects that fail fastest start from an attractive demonstration. A demo works on ten chosen examples; in production, ambiguous cases, out-of-scope requests and poor-quality data quickly appear. The pilot is precisely there to find them before investing. ## Safeguards and delegation The more AI produces customer-facing content, the more control is needed. A reply sent without review, an automatic financial recommendation or an HR decision made by a model are unacceptable risks without explicit human supervision. ## Measure before scaling up A successful pilot is characterised by three elements: measured time saved, result quality at least equal to the manual version and real adoption by teams, not just by management. Without these three conditions, extending the project would only multiply the cost.When this guide is for you
- You hear about AI everywhere but nobody knows what to do concretely in your company
- Your teams already use AI tools without rules, risking leaks of sensitive data
- You want to reduce time spent on repetitive tasks (emails, quotes, reports, customer service)
- You are preparing an AI project and need to choose between ready-made tools and custom development
- Management expects an AI budget and you need to present costed use cases
Step by step
- Task mapping — List repetitive, time-consuming tasks by department, estimate the time spent and the annual cost. This inventory avoids starting with technology rather than need.
- Assessment and prioritisation — Rank use cases by estimated gain, implementation difficulty and risk (sensitive data, compliance requirements). Keep two or three realistic cases for a pilot.
- Measured pilot — Deploy the priority use case on a limited scope, with success indicators defined from the outset. Measure time saved, result quality and real adoption rate.
- Safeguards and training — Establish a usage policy: which tools, which data is allowed, which human checks before publishing. Train teams to verify answers, because a model gets things wrong with confidence.
- Scaling up — Extend to other teams, automate workflows and track costs (subscriptions, API calls, checking time). Plan a review every six months to adjust the tools.
What it costs (in MAD)
| Service | Indicative price | Good to know |
|---|---|---|
| AI assessment (mapping and priorities) | 8,000 – 40,000 MAD (≈ 730 – 3,700 €) | Depending on the number of departments and the size of the company |
| Pilot project in production | 15,000 – 80,000 MAD (≈ 1,400 – 7,300 €) | One use case, with integration and measurement |
| Assistant subscriptions (ChatGPT, Claude, copilots) | 200 – 600 MAD / user / month (≈ 20 – 55 €) | Professional plans to protect data |
| Developing an AI application | 60,000 – 250,000 MAD (≈ 5,500 – 23,000 €) | Only after a conclusive pilot |
| Team training | 4,000 – 15,000 MAD (≈ 360 – 1,400 €) | One to three days depending on the number of participants |
| AI API usage cost | 0.5 – 20 MAD / 1,000 calls (≈ 0.05 – 1.8 €) | Highly variable depending on the model and document size |
Indicative ranges for the Moroccan market in 2026, excluding tax. Prices vary with scope and provider.
Mistakes to avoid
- Launching an AI project without measuring the time actually spent on the task before AI
- Letting teams paste customer or financial data into consumer tools
- Entrusting sensitive decisions (recruitment, credit, sanctions) to a model without human oversight
- Choosing a tool first and then looking for a problem to apply it to
- Believing AI replaces a missing skill: an assistant speeds up an expert, it does not create one
Frequently asked questions
Where to start with AI in an SME?
With the repetitive tasks that take time for several people: drafting replies, meeting summaries, preparing quotes, sorting messages. A few days of scoping is enough to identify two or three testable cases.
Is my company's data safe?
Only with suitable tools and settings. The professional plans of major assistants exclude training on your data and offer controls. Free consumer accounts do not provide these guarantees.
How much does an AI project really cost?
A serious pilot often costs between 15,000 and 80,000 MAD, a large part of it in integration and verification. Subscriptions are low, but usage becomes costly at scale and human supervision remains essential.
Can you do AI without technical skills?
You can achieve real gains with ready-made tools. However, to integrate AI into internal systems, protect data and maintain solutions, technical skills or a provider are still needed.
How do you know whether AI is working?
By measuring three things before and after: time spent on the task, quality of checked results and correction rate. Without these measures, adoption rests on no evidence.
Updated on
Related guides
Version française: /services/strategie-digitale/conseil-ia/ · AR · NL