If your catalogue runs into the thousands of SKUs, generic AI advice doesn't hold up. A chatbot that works beautifully for a boutique store with 40 products falls over fast when a shopper is trying to find one specific item among 15,000 variants, sizes, and configurations.
Retailers running large, complex catalogues are under real pressure right now. Basket sizes are shrinking as shoppers spread their spend across more retailers and hunt harder for value, which means every visit has to work harder to convert.
At the same time, expectations have shifted. The majority of shoppers who've tried AI-powered product search now say it's become their default way to shop, and they expect the same relevance and speed regardless of how big or messy the underlying catalogue is.
So what actually separates an AI agent that works at scale from one that quietly becomes another support headache? Here's what we look at when we're helping high-SKU retailers get this right.
Why catalogue size changes the problem entirely
A small catalogue can get away with basic keyword matching and a handful of manual merchandising rules. A large one can't.
Once you're managing thousands of SKUs across multiple categories, variants, and stock locations, the core challenge shifts from "can the AI answer a question" to "can the AI actually understand intent inside a catalogue too large for any human merchandiser to hold in their head."
What to actually evaluate
What this looks like in practice
The pattern is consistent: the agents that perform well are the ones with a clear testing phase before full rollout, and backed by ongoing monitoring rather than a "set and forget" launch.
That's the deployment process we follow with every client, plan and configure against real data, test and launch in a controlled way, then optimise and scale once we can see what's actually working.
It's also worth being honest about where this goes wrong. The most common failure mode we see isn't a bad AI model, it's a good model connected to incomplete or outdated data.
An agent that can't see your real-time stock position will confidently recommend products that are out of stock.
The takeaway for high-SKU retailers:
The AI landscape has matured well past "does the retailer have a chatbot." The real question for 2026 is whether that AI can hold up against the actual size, mess, and speed of change in your catalogue.
Sources
- McKinsey & Company, Unlocking the next frontier of personalized marketing: https://www.mckinsey.com
- Triple Whale, AI in Ecommerce Statistics 2026: https://www.triplewhale.com/blog/ai-in-ecommerce-statistics
- Envive, 63 AI Personalization in eCommerce Lift Statistics: https://www.envive.ai/post/ai-personalization-in-ecommerce-lift-statistics
- Australia Post eCommerce Report 2026: https://auspost.com.au/business/ecommerce/ecommerce-report
