Optimizing Display: Lessons from 'Dead' Shelves in Cold Storage

09/09/2026

Many still believe that display optimization is solely the domain of brand image teams. They assume that arranging products by color, logo, or regional management directives is sufficient. However, implementation reality proves this completely wrong. Shelves may look beautiful but attract no buyers, or buyers may have to wander aimlessly to find what they need. The issue is not aesthetics, but whether the product is in the right position at the exact moment demand spikes. This is a retail analytics challenge where real-time sales data must directly interact with physical space.

The Problem of 'Sleeping' Square Meters

I recall a retail chain in the South specializing in fast-moving consumer goods (FMCG). Their challenge seemed simple: sales per square meter were declining, while real estate costs rose steadily each year. They hired a major consulting firm to redesign their planograms. The result was beautiful blueprints, but after implementation, revenue remained stagnant. Why? The designs were static. They were based on quarterly average data, while Vietnamese consumer buying behavior changes rapidly with seasons, promotions, and even weather.

When speaking with operations directors, they often complain that cashiers and sales staff cannot keep up with the data. They sell based on intuition and experience. A skilled salesperson might 'feel' that high humidity today means higher demand for moisturizing products, but the inventory management system remains unaware. This disconnect between floor reality and computer data is the gap modern AI solutions target. We don't need a system to replace humans; we need a system that helps humans make faster, more accurate decisions.

Real-World Conditions: Dirty Data and Rigid Processes

Before discussing algorithms, we must address the harsh real-world conditions in Vietnamese factories and warehouses, as well as in neighboring markets like Thailand and the Philippines. Sales data is often noisy. Products are scanned incorrectly, offline promotions are not entered into POS systems, or stockouts occur without alerts. If you feed 'dirty' data into an algorithm, you will get 'crazy' results.

Another obstacle is the decision-making process. In many enterprises, changing the position of a bottle of motor oil or a pack of instant noodles on a shelf requires at least three levels of approval. Meanwhile, real-time data shows Product A is selling twice the forecast, yet Product B occupies its space. If you must wait three days for permission to move stock, the opportunity is lost. The challenge here is not just technology, but breaking the cycle of traditional approvals. We have implemented solutions with major corporations like Masan and partners in the lubricant industry, and found that the biggest barrier is not software, but people's readiness to trust a data signal over 'tradition'.

The Algorithmic Fork: Choosing Between Speed and Accuracy

When tackling this retail strategy challenge, we face two major paths. The first is using static sorting algorithms. These are easy to deploy and low-cost, but they only work well in stable markets. The second is using Agentic AI combined with Computer Vision. This is a harder path, requiring investment in camera infrastructure and big data processing, but it allows the system to automatically detect discrepancies between the designed planogram and shelf reality.

We choose the second path for cases with high product turnover. The system analyzes sales data within 15 minutes. If a product shows a sudden growth trend, the algorithm proposes a new position. The key here is 'explainability'. Staff don't need to know how the algorithm works, but they need to know *why* a product should be moved up. If AI simply says 'Move it,' staff won't act. AI must say 'Move it to eye level because this product's revenue increased by 20% compared to yesterday.' This transparency is the key to keeping humans engaged in the process.

What to Measure When You Can't Measure Everything

After implementation, we should not chase absolute numbers like '5% revenue increase.' Instead, we need to look at qualitative and behavioral metrics. Observe whether the average time for a staff member to find and rearrange stock has decreased. Previously, they spent entire mornings counting inventory and organizing. Now, thanks to automatic alerts from image recognition systems, they only need to handle anomalies. Labor efficiency increases, but more importantly, so does proactivity.

There is an interesting detail I observed in a project in Mexico, where multinational retail chains operate. When the system began proposing small, continuous changes, the sales team started monitoring the dashboard themselves. They no longer waited for top-down directives. They looked at the data and proposed ideas themselves. This is data culture. It doesn't come from installing software, but from people beginning to understand the value of information. Sales per square meter increase not because AI is magical, but because decisions are made faster and based on facts rather than intuition. AIVISION has accompanied this process in building the foundation, helping businesses turn raw data into specific actions, but ultimate success still belongs to the business's own operational capability.

If I Could Start Over, What Would I Choose Differently?

If I had the chance to start over, I would spend more time on human training. Technology accounts for only 30% of success. The remaining 70% lies in whether staff trust and use the tool. I have seen very expensive AI systems abandoned in the corner of the control room simply because users felt it made them look bad. They didn't want to be 'directed' by a computer. If I started over, I would design the interface so AI acts as an assistant, not a boss. It would present options, and humans would be the ones to press the confirmation button.

Ultimately, display optimization is not a one-time project. It is a continuous process. Markets change, products change, and consumers change. The question for any manager reading this article is not 'What can AI do for me?', but 'Is my organization flexible enough to change a product's position in 15 minutes without asking the boss?'. If the answer is no, then no matter how much money you invest in technology, you are still stuck in the past.

If you are weighing up a similar project, our team can help you scope it before you spend anything. See what we build or book a conversation.