Demand Forecasting

New Items, Stores and Channels

Forecasts built bottom-up from product and location attributes, down to each store and channel. For long lifecycle items, they pick up sparse demand, price and promo elasticity without additional work. For fashion and short lifecycle product, they work before you buy, for a whole collection, with no like items to pick. Forecasts feed assortment, buying and first allocation, and correct themselves continuously as real sales arrive.

Promotional Forecasting

Uplift for each promotion mechanic — price cuts, multi-buys, coupons, displays and more — alone and in combination. Halo and cannibalisation are modelled, even for promotions you have never run by grouping together history from similar promotions and similar product/locations, automatically with AI. Promotions are planned and optimised in Pricing & Promotions.

Lifecycle Phases

Every item, store and week is automatically tagged as pre-launch, launch, continuation, exit or cutoff using your existing sales, price and inventory data. Allocation and replenishment settings can follow the phase giving you fine-grained control over product tactics.

Automated Size Curves

Built end to end from messy size data: size ranges and sub-ranges are found automatically, curves are calculated per store and channel, and they update as sales come in. Size issues are automatically flagged and escalated. They feed assortment and allocation and drive prepack optimization in purchasing.

Slow and Sparse Demand

Dataviva Demand Forecasting leverages machine-learning methods to accurately predict demand for very low rate-of-sale items, such as jewelry, spare parts and pharmacy lines using dynamic aggregation across product and location attributes.

Explainable by Design

Dataviva chooses which product and location attributes drive each part of demand — rate of sale, seasonality, price response — and shows its choice. Forecasts approve themselves unless they need a look, and review goes to the lines that matter: high-margin items, traffic builders, or items where manual changes have stopped improving accuracy. Accuracy is tracked automatically, and out-of-stocks, outliers and unusual days are cleaned out of history before forecasting. The forecast also learns from its own mistakes.

Built for Tier-1 Volume

Horizontally scalable to millions of SKUs and thousands of stores, with full high availability and disaster recovery. Dataviva runs on top of your existing ERP and planning systems, so there is no rip-and-replace project.

Open to Your Data Science Team

Add your own methods using standard data science tools, call other systems in both interactive and headless processing and use forecasts as a service in any other system. See the Dataviva platform.