The Best Replenishment System Might Be the One You Already Have — Just Feed It Better Data


As our post-pandemic economy continues to take shape, retailers across the spectrum continue their struggle with a common bottleneck — inconsistent replenishment — leaving chronic bare spots on store shelves and customers frustrated. 

Simple math tells the story: A hypothetical retailer with 1,000 stores may carry as many as 10,000 SKUs at each location. In turn, those numbers could multiply into 10 million distinct decisions toward right-sizing inventory and allocation, in sync with localized customer demand. How can retailers achieve profit-optimal “right numbers” across every SKU? In short, which stores get how much, and how often?

Artificial Intelligence (AI) and machine learning (ML) have taken a lead role in the solution. This advanced data science is now rapidly refining and leveraging new dimensions of real-time, granular-level data — accurately forecasting effects of promotions and markdowns, while incorporating external factors such as projected shipping delays or even local weather patterns. 

This delivers new levels of hyper-targeted, data-driven replenishment and inventory optimization, reducing carrying costs of misallocated products, at the same time eliminating lost sales from extended out-of-stocks — and ultimately enhancing overall profitability.

Some C-suite retail decision makers have been wary of adopting AI/ML-powered solutions because they fear it might entail a full-scale, capex-heavy “rip-and-replace” of their entire replenishment infrastructure. In reality, those legacy systems may simply be hindered by a few “clogged arteries'' that would benefit from an improved diet of fresh data.  

The science behind the solution combines AI/ML algorithms with an automated technique for achieving an optimized balance between “short” and “long” risks — what data scientists have termed stochastic optimization. 

We can explain this process by comparing it to how a self-driving vehicle navigates roadways: An initial layer of AI/ML first “sees the world” by interpreting all available data points — recognizing differences between an ice cream truck and an ambulance — and assesses their likely paths. That real-time data is then filtered through an optimization engine that calculates the most efficient route between points A and B. 

Applied to retail replenishment, the base AI/ML layer interprets all available data variables — recent sales history, possible impact from promotions, and other pricing changes, along with supply chain disruptions or other external factors beyond the retailer’s control. That data is next fed through a stochastic optimization engine, which factors in all applicable business rules and other constraints.

Image Technologies
Credit: Technologies

The actionable results are data-driven replenishment cycles, focusing on a profit-optimized target inventory position, or what we call the order up-to point (OUTP) for every individual product across every store location. This, in turn, determines a suggested order quantity (SOQ), guiding retailers as they work with vendors to maintain shipments in the wake of further supply or logistics issues. 

Not so long ago, the very idea of achieving this scale of hyper-targeted, profit-optimized data intelligence for every SKU would have envisioned a small army of human planners perpetually wrangling with manual spreadsheets. Replenishment is just one example of how advanced AI has already proven to be a game-changer for modern retail. 

Sivakumar Lakshmanan, Head of the AI Forecasting & Supply Chain, and Stefano Alberti, Director of AI,, now part of Zebra Technologies