Idea
Zero-shot retail theft detection platform cutting false alarms and costs with model-agnostic AI orchestration.
Research Paper
Core Innovation
This paper introduces Paza, a zero-shot retail theft detection framework that orchestrates multiple existing vision models in a layered pipeline to minimize expensive model calls. It uses behavioral pre-filters to reduce costly vision-language model invocations by 240x, enabling cost-effective, scalable, and model-agnostic theft detection without any custom training.
Why It Matters
Retail theft causes massive economic losses and current AI solutions are costly and require extensive training. This platform reduces false alarms and operational costs by leveraging existing models without retraining, enabling scalable deployment across multiple stores with a single GPU. It also preserves privacy by obfuscating faces, addressing regulatory concerns.
Market Size (TAM)
$20–50B TAM for retail security AI; $2–5B SAM from retail chains and security providers. Driven by rising retail theft losses and demand for affordable AI surveillance.
Potential Customers & Pain Points
- Retail chains – High theft losses and expensive AI solutions
- Small to medium retailers – Limited budgets for theft detection
- Security service providers – Need scalable cost-effective AI tools
- Shopping malls – Require privacy-compliant surveillance systems.
Business Model
Subscription-based SaaS charging $50-100 per store per month, scalable by GPU usage and number of stores served, with optional premium support and model upgrades.
Competitive Landscape
- Standard AI theft detection vendors
- Amazon Rekognition
- AnyVision
- BriefCam
Implementation Challenges
- Integration with diverse retail camera systems
- Real-world variability beyond controlled datasets
- Customer trust in zero-shot detection accuracy
- Privacy regulation compliance across regions
Validation Strategy
- Pilot deployments in multiple retail stores to measure real-world detection accuracy and false alarm rates
- Cost-benefit analysis comparing existing solutions and Paza
- User feedback on privacy features and operational integration
- Continuous benchmarking against new vision-language models
Research Paper Overview
Zero-Shot Retail Theft Detection via Orchestrated Vision Models: A Model-Agnostic, Cost-Effective Alternative to Trained Single-Model Systems
Summary
Retail theft costs the global economy over $100 billion annually, yet existing AI-based detection systems require expensive custom model training on proprietary datasets and charge $200-500/month per store. We present Paza, a zero-shot retail theft detection framework that achieves practical concealment detection without training any model. Our approach orchestrates multiple existing models in a layered pipeline - cheap object detection and pose estimation running continuously, with an expensive vision-language model (VLM) invoked only when behavioral pre-filters trigger. A multi-signal suspicion pre-filter (requiring dwell time plus at least one behavioral signal) reduces VLM invocations by 240x compared to per-frame analysis, bounding calls to <=10/minute and enabling a single GPU to serve 10-20 stores. The architecture is model-agnostic: the VLM component accepts any OpenAI-compatible endpoint, enabling operators to swap between models such as Gemma 4, Qwen3.5-Omni, GPT-4o, or future releases without code changes - ensuring the system improves as the VLM landscape evolves. We evaluate the VLM component on the DCSASS synthesized shoplifting dataset (169 clips, controlled environment), achieving 89.5% precision and 92.8% specificity at 59.3% recall zero-shot - where the recall gap is attributable to sparse frame sampling in offline evaluation rather than VLM reasoning failures, as precision and specificity are the operationally critical metrics determining false alarm rates. We present a detailed cost model showing viability at $50-100/month per store (3-10x cheaper than commercial alternatives), and introduce a privacy-preserving design that obfuscates faces in the detection pipeline. The source code is available at https://github.com/xHaileab/Paza-AI.