Demand forecasting by product and location, with purchase order recommendations
We build demand forecasting for retail chains, pharmacies, restaurants and online stores. The model forecasts sales for every product and location weeks ahead. Buyers get a ready order suggestion and see why it is that number.
What hurts
Purchasing runs on "last year plus a bit". Some stores sit on stock for months while others sell out by Friday. Cash is frozen in inventory and customers walk to a competitor.
Your buyer keeps it all in their head and in a spreadsheet. They know Easter needs more and a heatwave needs less. When they go on leave or quit, that knowledge leaves with them.
Promotions, new stores and assortment changes break any simple formula. Write-offs of fresh goods grow, and nobody can explain where exactly the volume was wrong.
What we build
- Analysis of sales, stock, write-offs and stockouts by SKU and location
- A forecasting model with seasonality, holidays, promotions and weather
- Separate methods for new products and slow movers
- Order recommendations that account for stock, lead time and pack sizes
- Export of recommendations into SAP, Odoo or your ERP
- A forecast accuracy dashboard by category and location
- An explanation for every recommendation: what drove the number
- Weekly automatic retraining on fresh data
What you get
- A running forecasting service on a refresh schedule
- Order recommendations inside your ERP
- A dashboard of forecast accuracy and deviations
- Model code and methodology documentation
- Training for buyers and category managers
How we measure the result
- Forecast error by category and location, week over week
- Share of days with zero stock on key products
- Volume of write-offs and excess inventory
- Share of recommendations accepted by buyers without changes
Systems and stack
Industries
Questions
How much does demand forecasting cost?
From $10,000. The price depends on the number of SKUs and locations, the quality of sales history and whether recommendations must flow back into your ERP. A 20-store chain with clean data costs less than 300 stores on several systems. We fix the estimate after discovery, once we have seen your real sales history.
How much history do you need?
Ideally a year or more of daily sales, so the model sees seasons and holidays. Stock levels too, because without them a stockout looks like low demand. A promo calendar and write-offs help. With less history we start simpler and show you where it is weak.
How accurate will the forecast be?
We do not promise numbers in advance, because accuracy depends on your market and data. First we measure how your current purchasing method performs. Then we compare the model on the same weeks. You see the difference before the model affects any order.
Will it replace our buyers?
No. The model gives a recommendation and an explanation, and a person decides. It does not know about supplier deals or a new mall opening nearby. Buyers adjust the number, and we analyse those edits to improve the model.
Which systems does it work with?
We read sales and stock from SAP, Odoo, Microsoft Dynamics, POS systems or your warehouse. Recommendations go back as draft purchase orders or an import file. Heavy reads run at night, so the ERP stays fast for your team. Data is processed under a GDPR DPA.
How is the forecast rolled out?
We first run it on some categories or stores and compare against the current process. Then we expand. The model retrains weekly on its own. Under an SLA we watch accuracy and react when the market shifts sharply.
Related services
More in this section
Describe the task. We reply within one business day, the first call is with the founder.
Describe the task. The first call is with the founder.
We reply within one business day. 30 minutes to tell you whether it is worth doing and what it costs.