The verdict in three sentences
For a ready-meals SME running 4 lines with use-by dates of 5 to 12 days and supermarket orders that swing by 30% from one week to the next, AI scheduling combining demand forecasting and constraint-based sequencing costs €25,000 to €60,000 excl. VAT. It cuts waste by 10 to 20% (unsold goods, downgrades, donations) and changeover and cleaning time by about a third. The choice between an off-the-shelf APS and a custom model mainly depends on how specific your allergen and cleaning constraints are.
What AI changes in the production plan
Today the planner works on a spreadsheet fed by history, announced promotions and experience. AI adds two components. Forecasting combines history per product and per retailer, the promotional calendar, weather (salads and hot dishes react strongly to temperature in Toulouse) and public holidays; it reduces forecast error from 30-35% to 18-22% over a 7-day horizon. Scheduling then computes the production sequence that respects real constraints: allergen order (gluten-free before gluten, nut-free before nuts), full cleanings, blast chiller capacity, crew availability and the remaining shelf life required by the retailer on receipt, often two thirds of the use-by date.
Off-the-shelf APS or custom model
| Criterion | Off-the-shelf APS (food industry) | Custom model |
|---|---|---|
| Setup cost | €30,000 to €80,000 excl. VAT | €25,000 to €60,000 excl. VAT |
| Yearly licence | €12,000 to €40,000 | €0, hosting €2,000 to €6,000 |
| Deployment time | 4 to 9 months | 3 to 5 months |
| Handling of allergen and cleaning constraints | Configurable, sometimes rigid | Modelled exactly |
| Demand forecasting | Standard module, often optional | Trained on your retailers and local weather |
| ERP and MES integration | Standard connectors for major ERPs | Built on your ERP |
| Scalability, new line | Included by the vendor | Adjustment needed, €3,000 to €8,000 |
| Dependency | Vendor | Supplier, code delivered |
An off-the-shelf APS fits well beyond 8 to 10 lines or across several sites. Below that, a custom model often proves cheaper over 5 years and fits your constraints better. These amounts are 2026 orders of magnitude.
Measurable gains
| Indicator | Before | After (12 months) | Estimated yearly value |
|---|---|---|---|
| Finished goods waste | 4.5% of revenue | 3.6 to 4.0% | €70,000 to €130,000 |
| Stock-outs and retailer penalties | 2.2% of lines | 1.2% | €25,000 to €40,000 |
| Weekly changeover time | 18 h | 12 h | 6 extra production hours |
| Weekly planning time | 12 h | 4 h | Planner focused on disruptions |
| Overtime | 450 h a year | 300 h a year | €4,000 to €6,000 |
| 7-day forecast error | 32% | 20% | More accurate raw material stocks |
For an SME with €15 million in revenue, a 0.5 to 0.9 point reduction in waste alone represents €75,000 to €135,000 a year.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
Mini case study
Julien, operations director of a 110-employee ready-meals SME near Toulouse (€14 million revenue, 4 lines), was losing 4.6% of output to unsold goods and downgrades, i.e. €644,000. Custom project: €48,000 excl. VAT, plus €4,000 a year of hosting. After a year, waste falls to 3.8%, i.e. €532,000: €112,000 saved. Retailer penalties drop by €28,000. Net first-year gain: 112,000 + 28,000 - 48,000 - 4,000 = €88,000, with payback in about 4 months once the model is in production.
FAQ
How much history does forecasting need? At least 18 months of sales per product and per retailer, ideally 3 years to capture seasonality. New products are forecast by analogy with similar ones.
Does AI replace the planner? No. It proposes a plan in a few minutes; the planner validates it, adjusts it to disruptions (breakdown, absence, missing material) and spends about 4 hours a week on planning instead of 12.
How are last-minute retailer orders handled? The plan is recalculated with each new order in under 2 minutes, protecting runs already started. The planner sees the impact on other orders before accepting.
Do we need an MES or sensors on the lines? Not to start: the ERP and production declarations are enough. An MES improves rate accuracy, adding 2 to 5% of capacity.
Can the project be funded? Regional industrial modernisation grants exist in Occitanie and can cover 20 to 40% of such a project, subject to conditions to check at filing time.
Let's scope your project. Share your number of lines, your product list and 18 months of sales history: we price AI scheduling (€25,000 to €60,000 excl. VAT) and the expected gains, with a one-line pilot in 10 weeks. 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.
