AI Visibility in Mexico: When Data Meets Store Reality

04/10/2026

Expectations vs. Reality on the Border

When we began discussing the deployment of AI Visibility and display scoring in Mexico, most meetings followed a familiar script. Stakeholders expected that simply installing AI cameras in small grocery stores in central regions would allow the system to automatically “understand” everything and deliver absolutely precise figures. These expectations were often very high, sometimes even idealized. The reality was quite different. We did not see a clean picture from the start. Instead, we encountered a host of issues regarding lighting, camera angles, and the chaotic arrangement of merchandise that no technical documentation could fully describe.

The first thing that forced our technical team to regroup was the cultural gap in the data. In Vietnam, when working with partners like Masan or Meat Deli, we were accustomed to a certain level of standardization in the supply chain. In Mexico, particularly in the mom-and-pop retail sector, product placement relies heavily on the store owner’s intuition. Some days the shelves are full; other days they are empty. If we had rigidly applied the scoring criteria from other markets, the AI system would have thrown errors constantly. This was not a software bug, but a failure of expectations.

Before, During, and After Camera Deployment

The deployment process did not follow a straight line. The “pre-deployment” phase took up the majority of the time and is the most easily overlooked. We spent several weeks collecting real-world image data from the actual points of sale in Mexico. The goal was not to train the model immediately, but to “read” the market. We discovered that about one-third of the images were compromised by direct sunlight hitting the display case glass. In Vietnam, this issue is less common due to different store architecture and installation habits. This forced us to fine-tune the image processing algorithms, particularly white balance and noise reduction, so the AI would not be “blinded” by harsh light.

In the “deployment” phase, the biggest challenge was network infrastructure. Internet connectivity in some parts of Mexico is not as stable as in major Southeast Asian cities. The solution we applied, similar to our work with lubricant industry businesses in Vietnam, was to enable devices to operate offline and synchronize data when a connection was available. The AI camera records events, stores them locally, and only sends them to the server when the connection is strong enough. This prevents data loss due to network drops, which is the number one enemy of remote monitoring solutions.

The “post-launch” phase is where the problem became interesting. The system began generating display scores. However, instead of just looking at the numbers, we focused on feedback from the field supervision teams. They reported that the AI helped them detect minor issues that humans often overlook, such as products not being placed according to commercial standards (planograms) yet still being visible to customers. This is the true value of AI Vision: it does not replace humans, but expands their vision. Image processing speed met our commitments, with latency under one second, allowing managers to review store conditions in near real-time rather than waiting for end-of-week reports.

What Can Be Measured and What Remains Unsolved

Measuring the effectiveness of this project was not based solely on algorithm accuracy. We tracked the “successful intervention rate.” In other words, when the system flagged a shelf as out of stock, did the sales team fix it? The results showed that thanks to early AI alerts, the time to detect errors was significantly reduced compared to previous manual methods. However, we must be direct: AI cannot solve every problem. There are situations where the camera cannot distinguish between an out-of-stock product and one obscured by an unrelated object. These cases still require human intervention or an additional verification process.

The key point is integrating AI Visibility into operational workflows. If there is only hardware and software without a change in how the team works, the results will not be sustainable. We have seen many businesses invest in technology but fail because they did not retrain their staff. In Mexico, we had to work closely with supervisors to ensure they understood the meaning of each score and knew how to act based on that data.

Trade-offs and Next Steps

If we were to start over, there are several things we would do differently. We would invest more in standardizing input data. Requiring store owners to adhere to specific photography or camera installation rules would significantly reduce system errors. However, this requires patience and careful negotiation; it cannot be imposed mechanically. Additionally, we would consider deploying AI Office to support office teams in analyzing reports from AI Visibility. Currently, aggregating data from thousands of stores is time-consuming. An AI assistant could summarize key issues and propose solutions directly in emails or Excel reports, reducing the burden on managers.

Expanding to other markets like the Philippines or Thailand will continue to face similar but distinct challenges. Each market is a unique puzzle of retail culture and infrastructure. The lesson learned is: do not view AI as a magic wand. View it as a tool that must be nurtured with high-quality data and tight operational processes. AIVISION has accompanied many Vietnamese businesses on this journey, and the experience accumulated from demanding markets like Mexico will be a crucial foundation for us to continue developing AI Vision solutions suitable for a global context. The road ahead is long, and humility before real-world data is always the key to going far.

Vtraks is AIVISION’s application that brings AI Camera, AI Visibility, and AI Office together in one place. Try it on the App Store, or read more articles by AIVISION.