Catching Empty Shelves Before They Cost a Sale
Thread AI Editorial · July 10, 2026
A consumer goods manufacturer keeps its products stocked in stores using its own sales reps, who visit each location and refill the shelves. When a shelf goes empty, the sale is lost, and at some retailers, the manufacturer also pays a penalty.
With Lemma, a Worker spots which shelves are about to run empty, checks what the warehouse can actually ship, and gives each rep a short list of what to restock first.
What Lemma delivers
Millions of dollars in annual recovery
From fewer penalties and lost sales.
Stockouts caught the same day
While there’s still time to restock instead of days later in a report.
Planners focused on real shortages
Since routine restocks go straight to reps.
The Challenge
Every morning, large retailers send the manufacturer yesterday’s sales, store by store. Each retailer uses its own file format and its own codes for stores and products, so everything has to be translated before it can be compared. Even then, that data sits apart from the sales history, promotion calendar, and warehouse records needed to act on it, so problems often surface in reports days after a shelf has emptied.
The team works out which products are about to sell out where, factors in upcoming promotions and holidays, and checks whether the warehouse has enough to restock. When several stores need the same product from the same warehouse, there isn’t always enough to go around, and it’s not always clear who should get it first.
Pulling it all together takes hours of spreadsheet work across the sales, planning, and supply teams, and some shelves end up empty in the meantime.
The Solution
A Lemma Worker runs every morning, combining retailer sales files with years of sales history, upcoming promotions, and warehouse records to flag products that are already out or about to run out at each store. An AI model checks for local events that could push demand higher, like a heat wave or a big game nearby, and fixed rules rank each store’s flagged products so the most urgent come first.
With this Worker, reps receive a clear list before their visit, including which products to restock and how much. When the warehouse can’t cover a restock, it goes to a supply planner instead, who sees why it’s short and can request more stock with one approval. Stockouts are now caught the same day, often before the shelf is empty.
Lemma Capabilities
Hybrid deterministic and AI reasoning.
Rules score every store-SKU for stockout risk against five years of sales history and an upcoming promo and holiday calendar. An AI agent then researches real-world demand drivers near each at-risk store, such as weather or local events, to refine the forecast. It can only raise the forecast on real evidence, never lower it, while coverage, ranking, and routing stay rules-driven.
Unified data across systems of record.
Retailer feeds arrive in different formats with their own store and product codes. The Worker normalizes them and maps them to internal IDs, so sales history, event calendars, and ERP inventory all line up.
Grounded in what can actually ship.
Forecasted demand is checked against distribution-center stock, committed orders, and inbound purchase orders. When stores share a distribution center, stock is allocated by urgency, and each restock is tagged as covered, covered by an inbound purchase order, or short.
Tiered action with a human in the loop.
Coverable restocks go straight to each rep as a prioritized list of the top three SKUs per store. Only true shortages reach a supply planner, who sees the reason and suggested fixes for each line and approves purchase-order requests in one step.
What’s Next
With a restock list ready before each visit, reps can spend their time refilling the shelves that matter most instead of checking every product, and planners can sort out shortages before they reach the store. Over time, the manufacturer can keep its fastest-moving products in stock at the retailers that hold it to in-stock requirements, protecting both the sale and the relationship.
Does this sound familiar?
Every morning starts with pulling reports from sources that don’t use the same codes.
By the time the data is cleaned up, the problem has already happened.
When supply is short, it’s not clear who should get it first.
The people who handle real problems are buried in routine ones.