The verdict in three sentences
A building materials distributor with 8 depots, 12,000 SKUs and 45 million MAD (about $4.9 million) of inventory almost always holds too much stock on slow movers and runs out on the items that matter. AI forecasting by product family and depot, connected to the ERP and producing purchase order proposals, makes it realistic to target 15 to 25% less stock and 30% fewer stockouts within 6 to 9 months. The 2026 budget sits between 250,000 and 600,000 MAD (about $27,000 to $65,000) for a 3 to 4 month project, often paying back in under six months. The figures come from a Casablanca distributor case; a Dubai distributor applies the same method, with budgets usually quoted in USD or AED.
Where the excess stock hides
The imbalance is not uniform. Heavy families (cement, steel) turn fast but weigh heavily in value; high-variety families (tiles, sanitaryware, hardware) concentrate dormant SKUs. Here is the 2026 order of magnitude observed at multi-depot distributors.
| Product family | Share of SKUs | Share of stock value | Current stockout rate | Target stock reduction |
|---|---|---|---|---|
| Cement and binders | 3% | 18% | 6% | -8 to -12% |
| Steel and rebar | 4% | 22% | 5% | -12 to -18% |
| Tiles and floor coverings | 30% | 20% | 9% | -20 to -30% |
| Sanitaryware and plumbing | 25% | 15% | 8% | -18 to -25% |
| Paint and hardware | 30% | 12% | 7% | -20 to -30% |
| Insulation and joinery | 8% | 13% | 10% | -15 to -25% |
| Overall | 100% | 100% | 7 to 8% | -15 to -25% |
The models use what the ERP already knows: 3 years of sales history per depot, actual supplier lead times, minimum order quantities. They add local seasonality (the Ramadan slowdown, summer construction peaks, the effect of large public projects) and past promotions.
2026 budget item by item
| Item | 2026 range (MAD, excl. tax) | Approx. USD |
|---|---|---|
| Scoping and cleaning 3 years of history | 30,000 to 60,000 | $3,300 to $6,500 |
| Forecasting models by family and depot | 80,000 to 200,000 | $8,700 to $21,700 |
| Purchase order and inter-depot transfer engine | 60,000 to 150,000 | $6,500 to $16,300 |
| ERP connection (read sales, write proposals) | 40,000 to 110,000 | $4,300 to $12,000 |
| Stockout and overstock dashboard | 25,000 to 50,000 | $2,700 to $5,400 |
| Pilot on 2 depots and buyer training | 15,000 to 30,000 | $1,600 to $3,300 |
| Total | 250,000 to 600,000 | $27,000 to $65,000 |
| Operations: compute, hosting, model monitoring (estimate) | 4,000 to 12,000 a month | $430 to $1,300 a month |
The low end corresponds to a well-structured ERP and family-level forecasts; the high end to SKU and depot-level forecasts with automatic transfers and a legacy ERP that has to be worked around.
Method and success factors
The project runs in four stages. Month 1: data audit, cleaning duplicate item codes and one-off sales (a single public tender distorts a forecast). Month 2: building and back-testing the models, targeting a forecast error of 15 to 25% per family per month. Month 3: pilot on 2 depots, where buyers approve or correct every proposal. Month 4: rollout to all 8 depots.
Adoption remains the key: the AI proposes, the buyer decides. A dashboard shows the reasoning behind each proposal (trend, seasonality, safety stock), which lifts the proposal acceptance rate from 50% at launch to 80 or 90% after three months.
Mini case study
Youssef, supply chain director at a building materials distributor in Casablanca, runs 8 depots, 12,000 SKUs and 45 million MAD of inventory for 380 million MAD (about $41 million) of revenue. His stockout rate hovers around 7.5%.
Need a professional website?
Kolonell builds websites that attract clients, optimized for the Sénégalese market. Free quote in 2 minutes.
After a 480,000 MAD (about $52,000) project, stock falls by 18% in nine months, i.e. 8.1 million MAD (about $880,000) of cash released. Calculation: with a 20% annual carrying cost (financing, storage, breakage, obsolescence), yearly savings reach 1.62 million MAD (about $176,000). Stockouts fall by 30%: if they used to cost 2% of revenue, i.e. 7.6 million MAD, the distributor recovers 2.28 million MAD of sales, about 410,000 MAD of gross margin at 18%. The project pays back in under 3 months of recurring gains.
FAQ
Do we need a new ERP for AI forecasting?
No. The solution reads sales and stock from the existing ERP and writes purchase proposals back. A legacy ERP without an API simply increases the connection item, up to 110,000 MAD (about $12,000).
How much history is needed?
Ideally 3 years of sales per depot to capture seasonality and Ramadan, which shifts by about 11 days every year. With only 18 months, accuracy drops by 5 to 10 points.
What forecast accuracy can we expect?
An average error of 15 to 25% per family per month is realistic, compared with 35 to 50% for manual forecasts or moving averages. At SKU and weekly level the error stays higher, which is why calculated safety stocks matter.
Do buyers lose control?
No, they approve every proposal. Order preparation time usually falls by 40 to 60%, which frees them to negotiate with suppliers.
How soon does stock come down?
The first effects appear 2 to 3 months after go-live, once overstocks are sold through. The 15 to 25% target is generally reached between months 6 and 9.
Let's scope your project. Tell us your number of depots, SKUs and your ERP: we will price the forecasting solution (range $27,000 to $65,000) and a 3 to 4 month schedule. 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.
