AI USE CASE
New Product Demand Forecasting Without History
Predict launch-week demand for new SKUs using analogous product data and market signals.
What it is
Machine learning models identify analogous historical products and external market signals to generate demand forecasts for new SKUs before any sales data exists. Retailers using cold-start forecasting typically reduce initial overstock and stockout rates by 20–35%, cutting markdown losses and improving shelf availability at launch. The approach also shortens the planning cycle by giving merchandising teams actionable quantity targets weeks earlier than manual benchmarking allows. Reliable results require a clean catalog of comparable past launches and access to category-level trend data.
Data you need
Historical sales data for comparable SKUs, product attribute catalog, and category-level market or trend signals (e.g., search trends, sell-through rates).
Required systems
- erp
- ecommerce platform
- data warehouse
Why it works
- Maintain a rich, standardised product attribute taxonomy so the similarity engine can find meaningful analogues.
- Embed forecast outputs directly into existing merchandising planning tools to drive adoption.
- Include external trend signals (search volume, social buzz, competitor pricing) to improve cold-start accuracy.
- Establish a launch post-mortem loop to retrain models with actual sell-through data after each new product cycle.
How this goes wrong
- Insufficient or inconsistent historical product attribute data makes analogous product matching unreliable.
- New product is genuinely novel with no comparable past launches, causing model predictions to be no better than random.
- Forecast outputs are ignored by buyers who distrust model recommendations without explainability.
- Poor data governance means the product catalog used for training has duplicate or mislabeled SKUs.
When NOT to do this
Do not deploy this when the retailer launches fewer than 20 new SKUs per year — the modelling overhead and data requirements will outweigh the planning benefit.
Vendors to consider
Sources
This use case is part of a larger Data & AI catalog built from 50+ enterprise transformation programs. Take the free diagnostic to see how it ranks against your specific context.