Prognozowanie popytu. Dlaczego magazyn płaci za błędną sprzedaż?

Demand forecasting doesn't end with a quiet sales slide. It ends in a warehouse, on a shelf, in an occupied location, with cash tied up in unclaimed goods. In this episode, I demonstrate a mechanism that recurs regularly in FMCG companies: sales declares demand, planning attempts to average it, production organizes batches, purchasing secures the material, and the warehouse receives the physical result of all the assumptions. If the sales forecast is an opinion, not a process, someone will pay for that opinion. Most often, the warehouse, production planning, purchasing, and the company owner will pay. The starting point is a simple scene from an S&OP meeting. The sales director presents the forecast for the next quarter. The demand line looks calm. Then the warehouse manager shows the stock levels. PLN 2,800,000 worth of unclaimed goods sits on the shelves, and nearby items are missing, which are being sold immediately. This isn't a problem of report aesthetics. It's a problem of the decisions that triggered purchasing, production, receiving, and storage. The most common mistake is that a forecast has an author but no owner. A salesperson enters volume, a sales manager wants to show growth, demand planning looks at history, production asks about batches, purchasing looks at supplier minimums, and the warehouse has to find space. Everyone sees the same number, but no one feels fully responsible for its outcome. When a forecast doesn't pan out, a classic exchange begins: the customer postponed the order, a promotion didn't materialize, the market changed, production had to round off the batch, the supplier had a minimum order quantity. Each statement may be partially true. However, the company still has inventory it doesn't need and shortages where demand was real. In this episode, I break down demand forecasting into specific management questions. Who can question a sales forecast before it turns into pallets? Does every major adjustment have a reason, an owner, and a review date? Does a salesperson see the cost of overshooting, or only the risk of stockouts? Are base sales separated from promotions, product launches, customer declarations, and one-time restocking? The second major topic is the mix. In many companies, the forecast looks good at the total level, but the warehouse operates at the SKU level. A product family can look correct, yet one variant is backlogged, another disappears from the shelf, a third is waiting for a promotion, and a fourth never received a confirmed order. The warehouse doesn't stock the average category. The warehouse stores the index, batch, expiration date, location, and logistics unit. If a company calculates the forecast at the total sales value level, and the cost of error is incurred at the item level, then the money is sitting in that gap. The third issue is promotions. Promotions aren't consistent demand. A customer declaration isn't an order. Initial channel replenishment isn't yet a regular rotation. If a company lumps all these signals into a single forecast, the sales history becomes messy. Then, a human or demand forecasting model sees the increase and treats it as the new norm. As a result, production runs a series, purchases secure raw materials, the warehouse receives goods, and a few weeks later, everyone explains that the market has changed. The fourth issue is the cost of forecast error. MAPE is a necessary metric, but a percentage alone doesn't tell management what actually happened in the operation. A warehouse doesn't pay percentages. It pays in locations, man-hours, transfers, expiration dates, lost availability, and cash. The same percentage error can be insignificant for one item and very expensive for another. Therefore, forecast errors must be translated into operational calculations: inventory value, backlog time, occupied inventory, write-off risk, handling costs, and the impact on rotating items. The fifth topic is technology. Demand forecasting models based on machine learning are now available to mid-sized companies. This is the good news. The bad news is that the model won't fix a messy process. If promotions aren't flagged, shortages look like a lack of demand, sales adjustments are unjustified, and SKU master data is inconsistent, the tool will only calculate the consequences of the mess more quickly. First, you need to establish sources of truth, accountability, and adjustment rules. Only then does the technology begin to work for the company. Chapters: 00:00 Sales Forecast - When a Warehouse Pays for a Premonition 02:13 Forecasting Without an Owner 10:59 The Error Isn't in the Total, It's in the Mix 19:03 Promotion Isn't Constant Demand 26:17 Forecast Errors Must Have a Price 33:00 A Model Won't Fix a Dirty Process 40:19 A Forecast Must Close the Loop Our Tools: https://locura.tech - An Ecosystem of Logistics Tools https://app.locura.tech - 3D Modeler for Warehouse Design https://pronostico.locura.tech - Pronostico for Demand Fo...