AI Product Recognition: 5 Essential Steps Before Starting a Pilot

Step 0: Understand the Need and Define Strategy

The first and most important question to consider before implementing product recognition is what expectations need to be met.

Clear expectations lead to well-defined KPIs, ensuring measurable success. Many FMCG companies assume that AI product image recognition retail execution can achieve 100% accuracy, but even in theory, that level is unattainable. A well-implemented solution typically delivers 95-97% accuracy, though its effectiveness heavily depends on data quality.

Employees need proper training on how to capture product images correctly, holding the camera at the right angle and avoiding obstructions. While a few percentage points of errors (local niche exceptions and edge cases) are inevitable, businesses should focus on optimizing overall implementation rather than chasing unrealistic perfection.

Another critical question is: Where does AI product recognition fit in your business ecosystem? For example, companies focused on B2B FMCG ecommerce can integrate AI image recognition with their eCommerce solutions to enhance retail execution and optimize sales operations via closer engagement directly with retailers. However, what kind of flow will be the best for your business or sales cycle has to be clear from the very beginning.

Before implementing AI-powered product recognition, consider how it integrates into the company’s existing operations:

• Does the company have existing systems that AI product recognition should integrate with? For example, SFA, CRM, or other platforms that support managing merchandising, sales representatives, technicians, or, in general, your field team.
• What kind of real-time support does the team require?
• Does the company require built-in or separate analytics, dashboards, or integrations with other business intelligence tools?
• What specific business challenges is the company addressing?
• Where should the focus be on a main shelf, secondary placements, or other areas?
• What is the core that should be monitored? Does that include only your SKUs or also competitors’ SKUs and / or brands?
• Shall the focus be only on SKUs or also on price, promotions, etc?

Visual product recognition with AI is not just an addon – it’s your gateway to a smarter, data-driven strategy. By defining its role and objectives early you ensure smoother implementation, higher adoption, and long-term success. More importantly, it unlocks invaluable data that traditional methods can’t capture with such precision, fueling future AI and ML-driven optimizations across your business. To see how AI and ML turn this data into real business impact, check out our insights on AI for CPG companies.

Step 1: Evaluate the Current Landscape

Before adopting or upgrading an AI-powered product recognition system, it’s essential to evaluate the current situation. Ask yourself: What are the main pain points? What expectations are realistic? How prepared is your team for this transition? How fast do we want to adopt it?

The most valuable insights often come from those who use the system daily. Engaging employees through surveys or interviews can help uncover recurring challenges and common pain points. For example, teams may believe that manual data collection is highly accurate, but real-world analysis often proves otherwise.

Numbers don’t lie. Instead of relying on assumptions, evaluate IR’s key performance indicators (KPIs) such as product and/or price recognition accuracy, average recognition time, and synchronization time.

Step 2: Define Requirements and KPIs

To fully leverage an AI product recognition application, defining clear business KPIs is essential. Many companies attempt to implement everything at once without a clear sales plan -tracking all SKUs on the shelf, including all competitors data, analyzing price tags including every promo tag focusing on 1% of exceptions from day one.. This broad approach often results in inefficiencies and poor execution.

Instead of trying to do everything at once, a step-by-step strategy ensures a more structured and successful rollout:

1. Start with core priorities, such as monitoring the share of a shelf or checking secondary placements.
2. Focus on SKU analysis first before adding pricing classification.
3. Focus on brand-level tracking of your competitors rather than SKU-level tracking for all brands and categories; or select only specific categories that you want to track on SKU level (the data from such tracking can be used in the future to enable SKU-level forecasting).
4. Pilot in limited mode rather than big bang.

Beyond implementation, it’s crucial to define success criteria for Image Recognition in the CPG market and establish measurable KPIs. At the start, avoid treating every inconsistency in quality or accuracy as an issue that must be fixed immediately. Instead, investigate further – determine how the deviation actually impacts your business.

Business too often hesitates to approve the solution that works perfectly in 99% of cases due to edge situations affecting just 1%. However, focusing on these rare scenarios too early can impact the progress or even stop the project – prioritizing core functionality first, focusing on the base is crucial for the successful implementation.

Step 3: Assemble the Right Team

The success of a pilot project depends not only on the technology itself but also on how the team perceives it and their commitment to making it work. People naturally resist change, especially when they don’t fully understand it. That’s why companies must clearly communicate the purpose of the initiative and how it directly benefits employees.

At the core of this transformation is AI product recognition technology, but its effectiveness depends heavily on the accuracy of the data provided – data collected by people. If the input data is high quality, the results will be meaningful and actionable and it will be clearly visible from day one of the project.

However, if the data is poor, even the best AI-powered system won’t perform effectively, resulting in a lack of confidence in the project, proposed solution, and Artificial Intelligence overall.

Beyond motivating employees, it’s essential to acknowledge and address their concerns. Change can be met with skepticism, especially when performance metrics and compensation are involved. For example, if AI-powered monitoring affects monthly bonuses, some employees may initially resist the change. In manual processes, data adjustments may have been possible to meet performance goals.

With AI-powered product recognition comes from captured insights, the process becomes fully objective and eliminates manual alterations. This shift can expose discrepancies.

For visual product recognition with AI to succeed, employees must understand how it benefits both them and the business. When they recognize the value of AI-driven improvements, they are far more likely to embrace the changes and integrate them into their daily operations.

Beyond Artificial Intelligence adoption, seamless integration between sales force automation for FMCG and computer vision for FMCG applications can optimize field operations, enhancing efficiency in merchandising and sales strategies.

Step 4: Define a List of Preferred Vendors

Now comes the crucial step – selecting the right AI product recognition vendor. Many companies promise 100% accuracy, but it’s unreal. So, how do you make the right choice?

When selecting a vendor for AI-powered product image recognition, it’s essential to look beyond just the technology and consider how it will be implemented and supported. Many providers use similar algorithms, so the real differentiator is execution. Ask yourself:

• Does the vendor provide a clear, step-by-step integration process?
• Will they support your team during deployment and beyond?
• Does the vendor fully own the solution, or is it built with the help of third parties?
• Is the solution flexible enough to support various channels and placements?
• Does the vendor have their own annotation team or do they rely on 3^rd parties?
• Does the vendor have right methodology in place to manage new SKUs and changes to those already trained within the model?
• How many Neural Networks they can enable within a single model?
• Can they operate online, offline or both?
• What devices are required to enable the technology – do I need to equip my team with high-end devices or can I run it on low-cost devices?
• What does the vendor’s roadmap for the solution look like?
• What do the vendor’s clients say about the solution?

It’s also important to assess the AI product recognition system in terms of implementation methodology, speed of results, system performance, and customer support. The right choice can transform this technology into an effective tool for your business. A poor choice, however, could turn it into just another failed initiative.

Step 5: Ensure High-Quality Input and Start the Proof of Concept

By now, it’s clear that high-quality data and a well-structured pilot launch are critical for successful AI image recognition implementation. But what exactly defines high-quality data?

Key elements of high-quality data:

• Comprehensive Product Catalog. Before the pilot begins, the client must provide a detailed product catalog with high-resolution images of each SKU. If your company is planning a label redesign, it’s best to implement Shelf Image Recognition after the update to avoid retraining the model.
• Real-World Store Shelf Photos. AI models require real-world images for effective training. For accurate classification, each product should appear at least 100 times in the training dataset. This doesn’t mean taking 100 separate images — if a product appears 20 times in one image, it counts toward the total. Of course, synthetic data can help here, but won’t get you to the desired 95%+ accuracy.
• Image Quality Matters. Poor-quality images can significantly impact accuracy and performance. To achieve the best results, ensure high-quality input data, as it directly impacts how well AI-powered classification and real-time analysis perform.

Remember that the quality of the output is only as good as the quality of the input.

Conclusion

Implementing AI product recognition technology streamlines operations, enhances efficiency, and improves retail execution. However, success depends on careful planning, high-quality data, and a well-prepared team.

Businesses must define their strategy, set clear KPIs, and follow a structured, step-by-step approach -from proof of concept to full-scale deployment – to ensure smooth adoption and minimize risks.

When executed effectively, AI-powered product recognition transforms business processes, enhances accuracy, and unlocks new growth opportunities. Moreover, leveraging end-to-end CPG software solutions can further strengthen Artificial Intelligence implementation, ensuring seamless integration across business processes and maximizing return on investment.