Demand Planning: Optimize Inventory or Cut Shipping Costs?

11/09/2026

Accuracy Expectations vs. Response Latency

Operations leaders often start meetings with a firm belief: if consumption forecasts are accurate enough, shipping costs will drop and stock will always be available. They assume the problem lies in the algorithms, a lack of point-of-sale data, or a planning team that isn't agile enough. Reality is quite different. When implementing demand planning systems in practice, we have found that the biggest barrier is not the accuracy of the numbers, but the latency in reacting to local changes. A 5% forecast error is acceptable if you have flexible adjustment mechanisms. However, if the replenishment process is rigid and follows a fixed cycle, that 5% error can turn into a stockout crisis at specific retail locations, triggering emergency orders at double the cost of regular shipping.

Our team at AIVISION has worked with major enterprises like Masan and TTN in supply chain management. The common lesson learned is: there is no single model that solves everything. Choosing the wrong approach to optimize inventory might save you on warehousing costs but burn through your logistics budget at the last mile. This article compares the three most common approaches, analyzes the cost of each choice, and suggests when to use which.

Three Approaches and Their Costs

Instead of chasing the latest technology, look at your company's data structure and decision-making speed. Below are three methods we frequently see in the Vietnamese market and neighboring regions like Thailand and the Philippines.

Approach 1: Centralized Forecasting at the Main Warehouse. This is the most traditional approach. All sales data from retail points are aggregated to a single location, and a central planning team decides the replenishment quantity for each warehouse. The biggest advantage is tight control and the ability to negotiate with suppliers due to large order volumes. However, the fatal flaw is the neglect of local nuances. A heavy rainstorm in Hanoi can suddenly spike demand for bottled water in the Cau Giay area, but the central warehouse in the South cannot react within 24 hours. As a result, those retail points must buy from competitors or nearby warehouses at higher prices, while incurring emergency shipping costs to compensate. This method suits product categories with long lifecycles and stable demand, but it will fail for fast-moving consumer goods like instant noodles or alcoholic beverages.

Approach 2: Decentralized Decision-Making for Retail Clusters. In this approach, the authority to decide replenishment quantities is delegated to regional managers or store clusters. Regional managers have the most practical data on local markets; they know which areas are hosting events, which are experiencing power outages or flooding. Flexibility improves significantly, and response times shrink from days to hours. The price to pay is inventory fragmentation. Clusters may forecast too optimistically, leading to local overstock at some points while others remain understocked. This requires an extremely fast and transparent internal transfer mechanism. Without a real-time tracking system, you will end up in a situation where you are sitting on excess inventory yet still have to buy more.

Approach 3: AI-Based Hybrid Model Using Local Data. This is the trend we are implementing with partners like Meat Deli and Gene Solutions. Instead of choosing between the two extremes above, this model uses machine learning algorithms to forecast demand at each retail point based on dozens of factors: weather, holidays, 90-day sales history, and data from similar stores within a 5km radius. The system automatically suggests replenishment quantities but retains a manual intervention mechanism for exceptional cases. Initial implementation costs are higher due to investments in data infrastructure and software integration. However, the long-term benefits lie in minimizing both cost types: warehousing costs, as inventory is automatically balanced, and emergency shipping costs, as forecasts are more accurate. This is the best fit for modern retail chains with a large number of stores and high complexity.

When to Choose What and What Really Matters

If your business is in a rapid expansion phase, with the number of retail points doubling every year, consider Approach 2. Decision speed is more important than absolute accuracy. A slightly wrong but quick decision is better than an accurate but late one. However, do not neglect building internal inventory control processes. You need a system that reports realistically on spoilage rates, expired goods, and stagnant inventory. Without this, decentralization will turn into chaos.

Conversely, if you are a manufacturer with a complex distribution chain involving multiple intermediary levels, Approach 3 is the most sustainable choice. Investing in an AI system is not just about buying software; it is about changing how operations are organized. You need to retrain your planning team to work with machine suggestions rather than fighting against them. We find that companies that succeed in this are usually those where the operations director is willing to accept the initial imperfections of the algorithm and patiently fine-tune the model during the first 3-6 months. Immediate application without a parallel running phase is the main cause of failure in supply chain digital transformation projects.

Insights on Hidden Costs

The biggest cost in inventory optimization is not in the shipping invoice, but in lost revenue due to stockouts at the exact moment customers have demand. A missed order at a retail point may be insignificant in value, but it impacts customer experience and long-term loyalty. Look at the total cost of ownership, not just logistics costs. If a new system reduces shipping costs by 10% but increases the local stockout rate by 5%, it is a failure. The ultimate goal is not to have the least inventory, but to have the right amount of stock, in the right place, at the right time. This is a delicate balance between risk and cost, and no mathematical formula can replace the deep understanding of the local market held by your operations team.

There is more here than one article can hold. Keep reading on the AIVISION blog, look at our display scoring solution, or get in touch.