The verdict in three sentences
An applied AI project costs, in 2026, between 5,000,000 and 15,000,000 FCFA (about 7,600-22,900 EUR) over 6 to 10 weeks, depending on the use case. The safest gains come from repetitive, measurable tasks: stock forecasting, customer scoring, ticket triage, document extraction. The usage cost (LLM API per token) is marginal against the hours saved when the scope is well chosen.
Where AI pays off concretely
Not everything is worth automating. Target high-volume tasks with clear rules and available data. Here are the most profitable use cases for an SME.
| Use case | Budget FCFA (2026) | Expected accuracy | Hours saved/month |
|---|---|---|---|
| Stock forecasting | 6,000,000 - 12,000,000 | 80-90 % | 30-50 |
| Customer scoring / follow-up | 5,000,000 - 10,000,000 | 75-85 % | 20-40 |
| Ticket triage | 5,000,000 - 9,000,000 | 85-92 % | 40-70 |
| Document extraction | 6,000,000 - 14,000,000 | 85-95 % | 40-80 |
| Email reply assistant | 5,000,000 - 11,000,000 | 80-90 % | 25-55 |
Maintenance: 300,000 to 700,000 FCFA/month (retraining, supervision, prompt and threshold tuning).
Custom project vs pay-as-you-go LLM API
An LLM API bills per token: tens to hundreds of thousands of FCFA/month depending on volume. But the raw API does nothing alone: it must be integrated with the business, the data and the controls.
| Criterion | Custom AI project | Raw LLM API (pay-as-you-go) |
|---|---|---|
| Upfront cost | 5 - 15 M FCFA | Near zero |
| Monthly usage cost | 300-700 k maint. + tokens | 50,000 - 400,000 FCFA tokens |
| Business integration | Included | To be built |
| Data & compliance | Scoped, controlled hosting | Secure it yourself |
| Accuracy on your data | Optimised, evaluated | Generalist |
| Supervision & guardrails | Included | To be built |
| Ownership | Code & pipeline delivered | Vendor dependency |
The right pattern combines both: a custom project that calls an LLM API, with your data, your controls and compliant hosting.
Mini case study
Aisha, owner of a 45-person trading SME in Dubai, handles 600 support tickets/month triaged manually by 2 agents, i.e. 60 h/month at an estimated internal cost of 300,000 FCFA/month. She invests 7,500,000 FCFA in automatic ticket triage plus attachment extraction, target accuracy 90 %, plus 450,000 FCFA/month (maintenance + tokens ~120,000 FCFA). Gain: 48 h/month freed, first-response time cut threefold. Processing saving: 2,880,000 FCFA/year; year-one outlay: 7,500,000 + 5,400,000 = 12,900,000 FCFA. Beyond hours, customer satisfaction and sales recovered via automatic follow-up push ROI under 18 months.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
FAQ
Which use case should we start with?
The one combining high volume, clear rules and clean data. Ticket triage and document extraction often offer the best gain/risk ratio.
Is our data protected?
The project scopes hosting and access per applicable regulations. Sensitive data can stay on controlled infrastructure.
What accuracy can we target?
Depending on the case, 75 to 95 %. We set a confidence threshold above which AI acts alone, below which a human validates.
How much do LLM API calls cost?
Roughly 50,000 to 400,000 FCFA/month depending on text volume. Usually marginal against the hours saved.
Do we need a lot of data to start?
Less than before. 2026 approaches achieve good results with a few hundred well-chosen examples, then improve with use.
Let's scope your project. Describe the task to automate, its monthly volume and your data constraints, and we will price an AI project in the 5 - 15 M FCFA range. 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.
