Product recognition APIs are transforming the retail and logistics industries by delivering powerful, reliable, and accurate identification technology. Achieving optimal results with these APIs requires careful testing and thoughtful preparation. Developers and testers often overlook simple yet crucial aspects, leading to suboptimal performance.

This comprehensive guide details how you can effectively use product recognition APIs in testing environments, ensuring maximum accuracy and performance.

Why Proper Testing Matters

Testing is normally the first step in evaluating how the API and checkout system will behave before presenting it to customers or stakeholders. Often, non-representative test configurations are used, and the results lead to confusion about how the API works and if it will perform in production. Effective testing replicates real-world scenarios closely, allowing APIs to function exactly as intended in production environments. Poor testing practices lead to recognition errors,  and tend to hold the project from proceeding, despite the testing not representing how the recognition API will behave in the production environment

Below, we'll explore common pitfalls and how you can avoid them to get the best results from your technology.

‍

Common Pitfalls to Avoid

Understanding and addressing common testing pitfalls will significantly enhance the effectiveness of product recognition APIs in your projects. Let's delve into each critical area:

‍

1. Prioritising Image Quality

High-quality images are central to reliable AI performance. Product recognition APIs depend on image clarity to correctly identify and classify products.

Poorly captured images—blurry, dim, or incorrectly coloured—are major culprits behind misclassifications. Here’s how to maintain superior image quality:

• Sharpness and Clarity: Always capture clear, focused images that accurately represent products as they appear at checkout.
  
  
• Optimal Lighting: Ensure lighting conditions closely mirror real-world retail settings. Images shouldn't be too bright or too dim.
  
  
• Consistent Colour Settings: Turn off auto white balance. A stable and consistent colour balance helps maintain accurate product identification.
  
  
• Consistency: Use lighting conditions that are similar to the production environment the API will be used in
  
  

Sample images are often available from API providers—reach out for access to these valuable resources.

‍

2. Selecting Appropriate Cameras and Devices

Camera choice dramatically affects API performance. Testing with inadequate devices leads to unreliable results. Selecting the correct equipment ensures consistent and accurate outcomes:

• Avoid Basic Webcams: Low-quality webcams, especially those in laptops, lack the necessary controls for accurate testing.
  
  
• Use Dedicated Cameras: "The best camera to use is one built for the purpose of the application. Cameras such as the Datalogic, Zebra and Tiliter cameras are best, since they are specifically built for this application.
  
  
• Stability and Positioning: Cameras should be placed in a stable position to ensure consistency in framing and angle. Tiliter provides extensive guidance on the correct positioning of the camera used with the API to ensure maximum performance.
  

‍

3. Correct Product Presentation

The way products are presented during testing significantly impacts recognition accuracy. The AI model is trained for specific presentation scenarios typical in retail settings. Deviating from these presentations introduces unnecessary confusion and inaccuracies. Here's how to correctly present products:

• Flat Surfaces: Products should always be placed on flat, stable surfaces similar to checkout counters.
  
  
• Avoid Human Interactions: Images should not include products being held by people. This scenario is significantly different from retail checkouts and can mislead the AI.
  
  
• Screens Are a No-Go: Avoid presenting images displayed on screens (phones or tablets). These are detected as potential fraud by the API and will be rejected.
  
  
• Full Visibility: Always capture the entire product clearly in the image. Partial or obstructed images reduce recognition accuracy.  Use single item types: The standard API setup does not handle multiple products at once. Since at the checkout most items are sold by weight, one product at a time is recommended.

‍

4. Use Real Products, Not Substitutes 

To ensure your product recognition API delivers accurate results, it's essential to test with real products—not replicas or placeholders. Artificial items like plastic fruits, decorative produce, or printed cut-outs do not accurately represent the size, texture, or coloration of real food items. These differences confuse the AI and often result in poor recognition performance. The system is also tuned to reject anything it perceives as potentially fraudulent. For example: 

• Plastic or decorative produce: Often lacks the surface texture and size variation of real items. Printed or cut-out images: Flat visuals distort depth and shape, leading to low confidence or rejection. 
• Images on screens: (as above) Phones or tablets used to display product images are detected as potential fraud and automatically rejected.

These safeguards are built into the API to protect against manipulation and ensure trustworthy use in real-world settings. For best results, always use fresh, real produce and present it in a way that reflects how it would appear during actual retail checkout.

‍
5. Accurate Product Configuration‍

Even perfect images won't deliver accurate results if the product configurations within the API are incorrect. Proper product setup is fundamental for accurate recognition. Common mistakes include incorrect or incomplete product listings, mismatched IDs, or missing categories. Prevent these issues by:

• Double-Checking Configurations: Verify that all product details are accurately configured, including mapping in your API account.
  
  
• Regular Updates: Periodically review and refresh your product catalogue. New products must be accurately listed and configured, and mapped before testing.
  
  
• New Products: If an archetype doesn't exist for your new product. Make sure the product is still created and apply the selection report when it is selected to provide the ability for the system to learn the new product over time.
  
  
• Prompt Configuration: As soon as new products are introduced, promptly configure and verify them to ensure smooth recognition from the start.
  
  

Conclusion

Testing your product recognition API with care and consistency leads to better results in production and speeds up the process when getting customer and stakeholder buy-in. It's not just about catching bugs—it's about building confidence in how the system performs in real-world environments.

Stick to the basics: high-quality images, correct device setup, proper product presentation, and accurate API configuration. These simple actions go a long way in helping your system respond faster, make fewer errors, and deliver a smoother experience.

If you're looking for a reliable partner in this space, Tiliter offers proven, enterprise-ready solutions that work across retail, logistics, and manufacturing. You get speed, accuracy, and a team that’s ready to support your goals.

‍