Global Toy Manufacturer Uses AI to Accelerate Retail Inventory Decisions

For a global toy manufacturer whose products appear in thousands of retail locations, keeping store displays stocked is an ongoing logistical challenge. Products are sold in grocery stores, airports, convenience stores, and other retail environments, each requiring regular inventory checks to determine when to order new shipments.

With such a large national footprint, the manufacturer relies on field workers to photograph store displays, which employees then manually review to determine whether inventory needs to be replenished. With thousands of photos coming in, the manual review process had become a bottleneck. It could take five to six days for an image to be reviewed and for a replenishment order to be placed, leaving displays understocked while inventory decisions waited in a queue.

The manufacturer partnered with SPR to explore how artificial intelligence could accelerate the process, improve the reliability of inventory decisions, and help employees focus on what mattered most.

The company managed the collection of inventory information by using field workers to photograph product displays at retail locations. Once collected, the photographs are made available to employees for manual review to determine whether inventory replenishment is necessary. 

Reviewing an image typically took four to five minutes. If additional inventory was needed, an employee made the determination, and another employee entered the decision into the company’s enterprise resource planning (ERP) system. Orders were then processed through the company’s existing warehouse fulfillment cycle. 

At the volume the manufacturer was handling, those few minutes added up quickly. Employees were spending significant time evaluating displays, while the backlog meant stores could wait nearly a week before a replenishment order was even initiated, potentially resulting in lost revenue for both the stores and the manufacturer. 

The process also posed challenges beyond speed. Submitted photographs sometimes had quality or consistency issues that required additional manual review, creating inefficiencies and potentially affecting the accuracy of inventory decisions. 

Inventory decisions also depended heavily on individual judgment. Different reviewers might assess the same display differently, and there was no consistent, documented framework for determining when replenishment was needed. 

The company needed a way to evaluate incoming images more efficiently while maintaining oversight of the ordering process. 

SPR designed and developed an AI-powered display intelligence system capable of analyzing retail photographs, identifying potential problems, and estimating inventory levels before presenting results to employees for review.

Rather than replacing the company’s existing inventory and fulfillment systems, SPR introduced AI-powered automation to streamline inventory analysis and accelerate replenishment decisions.

The solution uses AI-powered processing to evaluate incoming photographs through four stages:

• Image validation: Assess incoming photographs for quality, relevance, and potential inconsistencies.
• Display classification: Identify and categorize retail displays to support appropriate inventory analysis.
• Image correlation: Analyze relationships between photographs to support accurate inventory assessments.
• Inventory estimation: Estimate inventory levels and apply configurable business rules to support replenishment decisions.

SPR also developed a review interface that displays AI-generated recommendations and supporting reasoning alongside the original photographs. Employees can evaluate the recommendations, make faster, more informed replenishment decisions, and escalate issues that require further attention.

This human-in-the-loop approach allows the manufacturer to introduce AI into an important operational process without immediately handing over ordering decisions to automation.

The solution was built around the manufacturer’s existing technology environment, with appropriate security and data privacy controls in place to protect company information.

SPR also designed the processing stages to be modular and configurable. Business rules and inventory thresholds can be adjusted for different displays, allowing the manufacturer to refine the system through configuration rather than rebuilding the underlying workflow.

The solution integrates with existing business systems and ERP processes, introducing AI-powered inventory analysis and decision support without disrupting the manufacturer's established operations.

SPR delivered the first phase of the display intelligence system in just four weeks. Using a specification-driven development approach, the team accelerated the build while working through challenges involving image recognition, display classification, and reliable inventory estimates.

Initial testing demonstrated strong accuracy across the system’s core functions:

• 97.4% display classification accuracy, allowing the AI to identify different types of retail displays.
• 94.9% image correlation accuracy, demonstrating the system's ability to accurately associate related photographs.
• 6.9 percentage-point mean absolute error in estimating display fullness, providing a close approximation of human assessments.
• 37 seconds average AI processing time per set of retail photographs, with analysis performed automatically in the background, allowing employees to review pre-analyzed results rather than spending four to five minutes manually evaluating photographs.

These results are especially significant given the variety of environments in which the manufacturer’s displays appear. For example, retail displays surrounded by colorful merchandise and visually complex backgrounds can be challenging for AI models to accurately identify and analyze. SPR refined the image-processing workflow and AI instructions to improve performance in these real-world conditions.

Beyond faster image analysis, the solution helps identify image quality issues and duplicate photographs before they affect ordering decisions. It also establishes a more consistent approach to assessing display fullness across locations.

Beyond accelerating inventory analysis, SPR designed the solution to support a progression toward greater automation. The initial implementation uses AI to generate inventory recommendations while keeping employees in control of replenishment decisions. As the manufacturer gains confidence in the system, configurable business rules can enable increasing levels of automation, allowing employees to focus on exceptions that require human judgment.

By combining AI-powered analysis, configurable business rules, and human oversight, SPR delivered a solution that improves the consistency and efficiency of inventory assessments today while providing a foundation for more autonomous inventory management in the future.