Turning 12,000 SKUs into a sellable commercial-intelligence product

A retailer carrying 12,000+ SKUs had accumulated a large volume of internal data that simply was not usable for commercial intelligence. Product decisions relied on guesswork, there was no unified view of competitor designs and offerings, and no strategy existed to monetise the insights locked inside the business.

They wanted to move from operational reporting to true commercial intelligence – and, if possible, turn that intelligence into a product they could sell.

We delivered an intelligence stack in two phases: a 3-month data preparation and annotation phase, followed by a 6–9 month implementation with ongoing enhancement.

Key features shipped: a Competitor Data Extraction & Normalization pipeline scraping and reconciling market data; a Product Trend & Hot-SKU Detection Engine to prioritise stocking and promotions; a Peak-Season Prediction Model to cut stockouts and carrying costs; and a Customer Search & Design Preference layer that exposes real-time demand patterns.

All of it rolled up into a Commercial Intelligence Dashboard designed from day one to be sellable as a standalone product. Tech stack: Databricks, Snowflake, Apache Airflow, MLflow, PyTorch, AWS.

40% improvement in seasonal-prediction accuracy and 40% fewer stockout incidents in peak seasons, with 100% real-time market insight reporting replacing lagging manual reports.

Lifted profit margin on SKUs by 4.8% and opened a new revenue stream by productising the Commercial Intelligence Dashboard – converting an internal project into an external product.