About Unilever
Unilever is one of the world’s largest consumer goods companies, with a portfolio spanning food, beverages, personal care and home care — including brands such as Dove, Lux and Clear. This project focused on Unilever’s nationwide, multi-echelon distribution network in China, which serves around 150 million households.
The challenge
Unilever wanted to lower inventory across the whole network without hurting service. That meant designing the right safety stock for every SKU at every distribution center — enough to protect availability, but no more than necessary.
Getting there required four things working together: a better way to forecast demand, a smarter production and deployment strategy, a rigorous inventory optimization method, and a simulation tool to prove that the resulting safety stock levels would actually hold up.
The solution
Convect delivered an end-to-end approach built on its Algoflow forecasting engine and its integrated planning model.
1. Forecasting the full distribution of demand
Algoflow handles Unilever’s broad product range and varied demand patterns by selecting the right model for each item automatically. Crucially, instead of forecasting a single expected value, it forecasts the distribution of demand — the realistic range of what could happen. That avoids a common failure in inventory optimization: setting safety stock on the basis of a wrong assumption about demand variability.
2. Network-wide production and deployment optimization
Rule-based production and deployment planning was replaced with an optimization model built on a multi-period network. The model optimizes resources across the whole supply chain — triggering transfers, replenishment, production and purchasing decisions — to meet as much demand as possible at the lowest cost for a given inventory policy.
3. Simulation to choose the right policy
Combining the demand distributions with the optimized deployment logic, Convect ran simulations to test and compare inventory policies and parameters. This let Unilever see how each policy would perform under realistic conditions before committing — a far more defensible way to set safety stock.
Results
- More than 10% lower total network inventory at the same service level, in simulation testing.
- Lower inventory holding cost and freed-up working capital.
- A repeatable, evidence-based method for setting inventory policy across the network.
The results validated both the Algoflow forecasting approach and Convect’s inventory optimization methods — and gave Unilever a more flexible, efficient way to run its distribution network.
Where agents come in
Inventory policy isn’t set once. With the forecasting and optimization foundation in place, Convect’s inventory monitoring agents can watch every SKU and location continuously, flag emerging stockout or overstock risk, and propose rebalancing or replenishment actions for approval.
