Aisle Wanderer

The Challenge

Standard search engines tolerate noise; if a user searches for "honey" and sees "honey shampoo" in the mix, they just ignore it. But for automatic cart substitution, precision is binary.

The "Repeat Cart" engine - intended to be a key driver for recurring revenue - was backfiring. Because existing algorithms matched keywords without understanding context, they frequently substituted fresh lemons with lemon-scented dish soap. This forced users to manually audit every single automated choice, defeating the entire purpose of convenience and causing them to abandon the workflow entirely.

The Solution

We moved beyond keyword matching by building a Dynamic Taxonomy Engine. While retail giants spend millions on manual data tagging, we automated this using LLMs to generate deep, context-aware tag clouds for millions of SKUs.

This system introduced "common sense" to the inventory - parsing the difference between an item's descriptive tags and its actual utility, automatically filtering out matches that are technically similar but practically irrelevant.

Results

• 94% User Acceptance Rate on automated substitutions (vs. 70% industry baseline).
• Revitalized the "Repeat Order" feature, turning a broken touchpoint into a reliable revenue driver.
• Automated Taxonomy: Achieved enterprise-grade categorization depth without the manual labor costs usually required to maintain it.