$2M saved every year.
Manual classification, gone.

Every product attribute needed a human to read it.

Classifying product attributes was a manual job. Every entry across the retail and eCommerce catalogue had to be read, categorised, and checked against a brand hierarchy by hand, a process that ate time and resource with every batch.

Manual review meant manual error. The more data came in, the harder it got to keep classification accurate and consistent, and the slower the firm could turn raw retail data into the insights clients were paying for.

Three constraints were non-negotiable:

• Manual classification, entry by entry
• Accuracy dependent on individual reviewers
• No way to scale with retail and eCommerce data volumes

A pipeline that reads the catalogue instead of a person.

The firm built a FastText-based pipeline to automate product attribute classification end to end, removing the manual review step entirely.

Machine learning models were trained to classify consumer panel reviews directly, cutting reliance on manual tagging. Underneath sat a hierarchical brand classification system, keeping the taxonomy accurate and easy to maintain as new brands and products were added.

The outcome is a data processing system built to scale with retail and eCommerce volumes rather than strain against them, with automation running from raw data through to classified output.

$2M in manual costs, gone every year.

Automating product attribute classification saved the firm $2M a year while making the underlying data more accurate and consistent. That reliability let the firm move faster for its own clients, modernising a core piece of market research operations and strengthening its competitive position in an industry that runs on trustworthy data.