Manage over 2 million SKUs with 5,000 new products updated daily? It's nearly impossible when you're relying on manual tagging. That’s a story of Lately, a marketplace platform connecting local mom-and-pop shops with wholesalers. Their extreme product image diversity posed a huge challenge for tagging. Images were pulled from countless sources. So, their quality was wildly inconsistent and unpolished. Their existing search solution couldn't keep up. They penalized Lately for high search volumes, leaving shoppers frustrated and product discovery painfully slow.

That's when Lately teamed up with YesPlz AI to build a custom solution designed for their unique challenge. YesPlz trained AI models specifically on noisy, real-world product images. A fully automated tagging system handles the entire 2 million+ SKU catalog in real-time, with zero manual effort. Lately now delivers personalized, relevant search experiences at enterprise scale, without volume penalties or slow discovery cycles.

Challenges

• 2 million SKUs and 5,000 daily updates made manual tagging impossible: With an enormous, constantly changing catalog, manual product tagging simply couldn't scale. The team was drowning in backlog. New products sat untagged for days, making them virtually invisible to shoppers. Even with a full team dedicated to tagging, they couldn't keep pace with 5,000 daily additions.

• Noisy, inconsistent product images disrupted AI tagging: Images came from numerous merchants and mom-and-pop stores, creating extreme variability. Many were unpolished, user-generated, or shot in inconsistent conditions. As a result, conventional AI systems trained on clean, studio-quality images failed. The noise level was too high, and accuracy suffered, making automated tagging feel impossible.

• Existing search penalized high volumes and couldn't scale: Lately's previous search solution charged more as search volumes grew and struggled to accommodate continuous SKU changes. The system wasn't built for enterprise-scale fashion retail. Shoppers experienced slow, irrelevant results. Product discovery cycles lagged behind catalog updates. This created a frustrating disconnect between what was in stock and what shoppers could actually find.

Solutions: 

• Custom AI models trained on real-world, noisy product images: YesPlz didn't just apply off-the-shelf AI. We built custom models specifically designed to interpret Lately's unpolished, inconsistent product images. By explicitly training on noisy, real-world photos, the exact kind Lately worked with daily, the AI learned to extract accurate attribute data despite image quality challenges. It differentiated between signal and noise, dramatically improving tagging accuracy even on user-generated content. 

• AI automates tagging for 2 million+ SKUs in real time: Lately's entire catalog is now automatically getting tagged through YesPlz AI's advanced computer vision technology. It eliminates manual work, processing 5,000 daily updates without delay. Every product receives accurate, consistent tags the moment it's uploaded. The system scales effortlessly, ensuring real-time catalog updates that keep pace with Lately's dynamic inventory.

• Enterprise-grade search built for massive catalogs without volume penalties: YesPlz AI delivered a scalable, customizable search ecosystem tailored specifically for enterprise fashion retailers. The solution supports high SKU volumes and continuous updates without penalizing search traffic. Shoppers now experience fast, personalized, relevant product discovery, no matter how large the catalog grows. The integrated system turned Lately's massive catalog from a liability into a competitive advantage.
  

Fashion Filtering: Before and After AI Tagging

Before

After

Automated Product Tagging Success With Noisy Images

Before

After

AI-Powered Curation: Before and After

Before

After

Lesson Learned

• Turning Messy Images into Clear Tags: Even noisy images can be recognized when AI models are trained correctly. Our system learned to differentiate signal from noise, dramatically improving tagging accuracy—even when product photos were far from polished. This experience taught us that the true power of AI isn’t just in handling neat, studio-quality images. Instead, robust training on messy, diverse data is what makes AI scalable and genuinely useful for enterprise fashion retailers managing vast, complex catalogs.

Behind the Curtain

• AI-Powered Data Automation for Large Online Marketplaces