The verdict in three sentences
A custom applied-AI business project costs MAD 60,000 to 180,000 (~EUR 5,700 to 17,100) after a 2-to-3-week scoping phase, at a local day rate of MAD 3,500 to 6,000. The golden rule: only launch if ROI is under 18 months, measured in hours saved, reduced error rate and added revenue. Monthly inference cost and maintenance of MAD 3,000 to 9,000/month must be built into the math from the start.
The method: scope before you code
Most AI failures come from a lack of scoping. Start by identifying ONE high-volume use case with clear rules, quantify the potential gain, then decide. Here are the stages and their budget weight.
| Stage | Duration | 2026 cost (MAD) |
|---|---|---|
| Scoping + feasibility | 2 to 3 weeks | 10,000 – 25,000 |
| Prototype on real data | 1 to 2 weeks | 15,000 – 40,000 |
| v1 development | 3 to 5 weeks | 35,000 – 115,000 |
| Go-live | 1 week | 8,000 – 20,000 |
| Monthly maintenance | ongoing | 3,000 – 9,000 / month |
If the prototype fails to prove a net gain within 18 months, stop: that is the best use of the scoping budget.
Model the return: three levers
AI project ROI is computed on observable variables, not on promises. Here is the grid for a document-processing or support case.
| Lever | Measure | Typical 2026 impact |
|---|---|---|
| Hours saved | h/week × hourly cost | 40-60 % of task time |
| Error rate | % errors × unit cost | −30 to −60 % |
| Added revenue | responsiveness, upsell | +5 to +12 % on the task |
| Inference cost | MAD/operation | often < 1 % of the gain |
| Recurring cost | monthly maintenance | MAD 3,000 – 9,000/month |
The decision to invest is made when the sum of annualized gains exceeds (project cost / 1.5) + annual recurring costs.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mini case study
Fatima runs a 28-person services SME in Dubai. A file-processing task consumes ~35 h/week at a loaded cost of MAD 90/h, i.e. MAD 3,150/week or ~MAD 151,000/year. An AI project at MAD 120,000 plus MAD 6,000/month maintenance (MAD 72,000/year) automates 50 % of the task: a ~MAD 75,500/year gain in hours, plus an error reduction valued at ~MAD 20,000/year, i.e. ~MAD 95,500/year in gains. Against MAD 72,000/year of recurring costs, the net annual gain is ~MAD 23,500, and the upfront investment is paid back in ~15 months — under the 18-month threshold, so green light.
FAQ
How do we avoid buying a useless AI gadget? By requiring a paid scoping phase (MAD 10,000 to 25,000) before any development, with a clear walk-away threshold if ROI exceeds 18 months. Scoping costs little and prevents a MAD 120,000 project with no return.
Which use case first? The highest-volume, most repetitive one with clear rules: document processing, tier-1 support, quoting. Avoid ambiguous or high legal-stakes cases in v1.
Can inference cost run away? Rarely if modeled at scoping: it usually represents less than 1 % of the gain. We set caps and consumption alerts as a safeguard.
Do we need an in-house data scientist? No for a well-scoped v1. The vendor delivers a usable system; maintenance (MAD 3,000 to 9,000/month) covers refinement. An internal business owner suffices.
What if ROI exceeds 18 months? We don't invest, or we shrink the scope to bring the return under the threshold. Better to walk away after a MAD 20,000 scoping than fail on a MAD 150,000 project.
Let's scope your project. Describe the task to automate, its weekly hour volume and an indicative budget (MAD 60,000 to 180,000): we'll scope feasibility and expected ROI. Detailed quote within 48 h. WhatsApp +221 77 596 93 33.
Mohamed Bah
Fondateur, Kolonell
Passionate about digital and entrepreneurship in Africa, Mohamed has been helping Sénégalese businesses with their digital transformation since 2020. Founder of Kolonell, he believes every SME deserves a professional and accessible online présence.
