Case Study: How a Fortune 100 CPG Company Went From a Manual Coding Ceiling to Full-Scale Product Catalog Coverage

Summary

‍

A Fortune 100 CPG company was coding its product file by hand, and coverage capped out near 68%. Harmonya stood up a Custom Product Catalog that codes 80,076 products every cycle at 98% accuracy, with 99.95% completeness on core attributes, delivered weekly at 4x the contracted cadence.

The Challenge

‍

A Fortune 100 CPG company needed every product in a large, constantly changing file coded to its own taxonomy so that insights, category, and commercial teams could all work from the same product truth. The problem was not the intent. It was the method. The file was coded by hand, and manual effort has a ceiling.

‍

Coverage capped out near 68% of products. With finite hours, the team triaged its effort to the products that mattered most and accepted that full attribution across the whole file was not feasible. That left real blind spots. In the most recent cycle, 26,185 products, roughly 32% of the file, arrived with no tags at all. Any analysis that touched those products was running on a partial view, and no one could say with confidence what was hiding in the untagged third.

‍

The manual cadence could not keep pace either. New products entered the file every cycle, and hand-coding could never catch up without pulling effort away from everything else. The team needed complete, accurate, current attribution across the entire file, and they needed it on a reliable schedule.

‍

The Solution

‍

Harmonya deployed a Custom Product Catalog: the client defines the attributes, and Harmonya populates and maintains them to spec. The engagement followed Harmonya's standard three-stage process.

‍

First, Harmonya analyzed the client's labeled data and, for every attribute, produced a clear definition, a full value profile with coverage and distribution, the exact source fields that carry the answer, and a recommended approach. Each attribute was then routed to the method that fit its logic, from deterministic rules to individual and multi-attribute prompts, and built one at a time with a human review gate before anything was saved.

‍

Because the model applies judgment over each product's own content rather than starting from what a person happened to tag, it codes the entire file the same way every time. That included the roughly 32% of products that had never been touched by hand. Nothing was left out of scope because it was inconvenient to reach.

‍

The Results

‍

The Custom Product Catalog turned a manual ceiling into full-scale, always-current coverage.

‍

• ‍80,076 products coded every cycle. The entire file is now attributed on a consistent schedule, not just the share that manual effort could reach.‍
• 99.95% completeness on core attributes, dollar-weighted. Coverage moved from a partial view to near-complete across the attributes that drive the business.‍
• 98% accuracy, independently validated. Quality is measured against an independent test set, not self-reported, so the numbers hold up in front of stakeholders.‍
• 26,185 previously untagged products, about 32% of the file, now fully coded. The blind spot is closed. Products that were invisible under the manual process are now part of every analysis.‍
• Delivered weekly, at 4x the contracted cadence. Fresh, consistent attribution ships on a predictable schedule, so the file stays current as new products arrive.

‍

By defining the attributes once and letting Harmonya maintain them, the team stopped choosing which products to leave behind. They replaced a manual process with a maintained Product Catalog that keeps the whole file accurate, complete, and current, and freed their people to work on the decisions that data enables rather than the data itself.