During our very worthwhile appearance at NRF earlier this year, quite a few questions from visitors to our booth revolved around the ongoing challenges affecting retail replenishment.
Nobody needs to belabor the problem—our post-pandemic economy is still plagued by unwieldy supply chain disruptions, leaving retail customers regularly disappointed with empty shelves and little choice but to consider competing stores (or online options) for everyday items they rely upon. From shoppers to your store managers to C-level leadership, I know frustration is everywhere.
Simply accommodating the numbers is more complicated than ever. If you’re a midsized national retailer with 1,000 stores, you may sell as many as 10,000 SKUs at each store—translating to 10 million periodic decisions to right-size inventory across every location. Who gets what, when and how often? Those are the multi-million-dollar questions, right? The key is to factor in the dynamics of lost unit sales – from that customer who makes an abrupt U-turn from a barren shelf – against the profit-eating carrying costs from slow-moving product languishing in warehouses and stockrooms. In fact, our conversations with retail decision-makers at NRF voiced a common refrain: ‘’We know we need to better leverage data” to sync replenishment with customer demand and achieve the elusive sweet spot for every SKU—minimized costs, maximized profits.
Many Vendors, Few Solutions
As in past years, that expo floor at "The Big Show” in New York was a veritable sea of vendors, many touting their own software products for tracking replenishment. However, these tools are commonly limited by three inherent drawbacks:
They operate around a broad, one-size-fits-all approach, typically with entire categories driven by arbitrary target service levels, while lacking advanced, granular-level capabilities for data-driven inventory optimization—maximizing the profitability of the inventory investment.
The drivers of those replenishment policies (target service level, weeks of supply, etc.) are not directly connected to the desired end-goal: profitability.
They may require an extensive—and usually capex-heavy—overhaul or “rip-and-replace” of a retailer’s existing replenishment system(s).
That’s why we always guide retailers to data-driven replenishment solutions that are based around a pragmatic blend of AI/ML data science and stochastic optimization—a technique that explicitly incorporates demand uncertainty into the decision-making process to determine the optimal balance between “short” and “long” risks.
This pairing of techniques is analogous to the science behind autonomous driving vehicles, where an AI/ML acquisition layer “sees the world” around the car, interprets it, and projects the future possible states—distinguishing an ice cream truck from an ambulance—and predicting their likely paths. This information is then fed to an optimization engine, which achieves the object of driving from A to B under all applicable constraints (rules of the road, safety, navigation efficiency, etc.)
Similarly, an AI/ML layer interprets all available data to predict demand, relationships among products, responses to promotions, price changes, and other causal factors as well as contingent ones such as weather, disruptions, etc. This data is then consumed by a stochastic optimization engine whose goal is to maximize the expected profit at the individual SKU/store/replenishment cycle level, under all applicable business rules and constraints.
For you, as a retailer, the actionable data output includes a profit-optimized target inventory position, or what we at antuit.ai and Zebra call the order up-to point (OUTP) for every product and store location, which in turn determines the suggested order quantity (SOQ).
The Combination of AI & Stochastic Optimization Explained