Walmart’s digital shelf labels have captured headlines and generated public opinion and regulatory concern over the controversial practice of dynamic pricing. (Walmart has said DSLs won’t be used for that purpose; it claims every customer in the store at a given time will get the same price.)
But this AI-powered tech is doing more than changing prices. It’s part of a broader system that links demand forecasting directly to inventory and pricing signals, allowing stores to react more quickly when demand shifts, and to adjust orders before gaps or overstocks develop.
At Kroger, the work is happening inside a different loop. Through its 84.51° data science arm, the company is feeding loyalty data, promotional response and customer behavior into forecasting systems that update continuously. That’s changing how stores order, how distribution centers allocate, and how promotions are executed.
That’s a much quieter revolution. When it comes to artificial intelligence in grocery, the biggest impact isn’t on flashy, impressive personalization or controversial dynamic pricing. It’s on forecasting.
And that, in turn, is changing operating models.
A New Opportunity With LLM Forecasting
Forecasting largely determines what gets ordered, where it goes, and how long it sits. That’s always been true. What’s changing is the precision and speed.
Consider fresh. A traditional system might forecast strawberry demand based on last year’s weekly averages. A newer model incorporates weather forecasts, local event calendars, and even school schedules. A warm weekend in the Northeast can now trigger higher allocations midweek, reducing both out-of-stocks and Monday-morning shrink.
That matters because fresh shrink typically runs in the 4% to 8% range, according to benchmarks from the Food Industry Association. Even modest improvements in forecast accuracy can translate directly into recovered margin.
Until recently, much of that depended on experienced store teams making judgment calls.
At Wakefern Food Corp., centralized procurement and the growth of private label programs like Bowl & Basket and Wholesome Pantry have increased the need for tighter alignment between demand signals and supply. More precise forecasting reduces variability across its member stores—particularly important in a cooperative system with diverse local demand patterns.
At Wegmans Food Markets, the pressure point is different but related. High-turn fresh and prepared foods departments leave little room for error; product moves quickly or it becomes shrink. Firms such as McKinsey & Company and RELEX Solutions estimate that AI-driven forecasting can improve accuracy by 10% to 20% in perishable categories. That’s enough to reduce spoilage while maintaining in-stock levels and protecting margin.
The same pattern is playing out in center store, just less visibly. If a national brand runs a deep promotion on pasta, older systems often overcorrect, flooding stores with product that lingers after the deal ends. Newer models are getting better at separating “true demand” from “deal-driven demand,” tightening orders and reducing backroom buildup.
In dense Northeast markets, that precision matters. A ShopRite in North Jersey and one in South Jersey may respond very differently to the same promotion. More granular forecasting allows allocations to reflect those differences instead of pushing uniform volumes across the system.
Adoption is accelerating. Research from firms like Gartner and Blue Yonder suggests AI-enabled demand planning is moving from pilot programs to scaled deployment across large retailers.
Operationally, the effects cascade.
Better Forecasting… But There Are Limits
Labor planning becomes more dynamic; stores can schedule around expected volume rather than react to it. Replenishment cycles shorten as forecasts update continuously instead of weekly. Promotional execution tightens, with fewer mismatches between what’s advertised and what’s actually on hand.
Retailers deploying these systems report incremental gains in on-shelf availability, often in the 1 to 2 percentage point range, alongside reductions in excess inventory, according to vendor case studies and industry analyses.
There are limits. Grocery data remains inconsistent, and promotions still distort demand signals. A poorly structured discount can look like a permanent demand shift if the system isn’t calibrated correctly. The result is improvement—but not uniform improvement. Tuning remains ongoing.
Even so, the direction is clear.
Forecasting is moving from a periodic planning function to a continuous process embedded in daily operations. That shift is much less visible than pricing or retail media, but much more foundational.
The implication is straightforward. The retailers that forecast more accurately will operate with lower cost, less waste and fewer missed sales. In a business where margins are narrow, that advantage compounds quickly—and quietly.
