Modeling accurate inventory
AI models engineered for uncertainty build a more accurate picture of reality
Most inventory systems rely on perpetual inventory, which drifts quickly and requires labor-intensive counts to keep orders accurate, especially when faced with the spoilage, mis-scans, and random weight items of fresh departments.
Afresh developed purpose-built machine learning models that account for uncertainty and perishability from the start to maintain more accurate inventory positions.
Key benefits
Handle uncertainty in demand
Captures the variability from seasonality, holidays, and weather, so recommendations hold up when conditions shift.
Forecast new items from day one
Afresh finds similar items already in your system, then generates a reliable forecast for the new SKU before its first day on the shelf.
Forecast every form an item sells
The same item often sells under multiple SKUs after in-store transformation. Afresh rolls all those forms up into one demand signal, so the forecast reflects total demand instead of fragmenting across SKUs.
Respond to trends
Incorporate trends and unexpected events into demand forecasts, allowing more agile adjustments in ordering when spikes or drops hit.
Better decisions start with accurate inventory
Every order, production plan, and DC buy starts with a position on what's actually on the shelf, in the backroom, and in transit. When that inventory position is wrong, grocers lose margin in several ways: high shrink and carrying costs on overstocked items, stockouts on items the system thought were available, and wasted labor on constant inventory checks and counts.
Traditional inventory logic doesn’t work in grocery
perpetual inventory (PI) is based on a simple mathematical approach designed for retail categories where data inputs stay clean and predictable. But PI breaks when it gets to fresh:
• Unrecorded shrink that most systems can’t see, especially at the item level
• Measurement errors that get carried forward (receiving items in cases and selling as eaches)
• Inventory drift that reduces accuracy and increases labor to fix it
Afresh built smarter logic for fresh—proven across the whole store
Afresh started in fresh because it was the highest-impact—and most underserved—area for grocers. Since it didn’t exist, we designed models to handle spoilage, random weight, in-store transformation, and inconsistent counts—and still determine accurate inventory positions.
How Afresh calculates inventory
01 — INPUT
Better data in, better estimates out
Creates a more reliable foundation for every forecast.
Afresh checks incoming data for quality, corrects inaccurate or mismatched values, and fills gaps where needed. The result is a cleaner picture of what is actually happening in each store.
02 — CALCULATION
Forecast the range, not just one number
Models what could sell—and how likely each outcome is.
Afresh assumes inventory data will never be perfect—especially in fresh—so it doesn’t rely on simple math like “shipments minus sales.” Instead, Afresh considers things like the quality of the sales and shipments data, factoring in likelihood of mis-scans, and the demand and perishability models to create a probability curve of what true inventory positions might be.
03 — DECISION
Turn the forecast into the best decision
Balances demand with inventory, freshness, and profitability.
These store- and item-level probability curves for demand, inventory, and perishability feed Afresh’s decision-making engine. The engine simulates orders across in-stocks, shrink, display sizes, and freshness, analyzing thousands of possible combinations to identify the order that best supports profitability.
How probabilistic demand forecasts power the Afresh platform
Store Orders
Afresh’s inventory model provides an estimate for each item, so order recommendations reflect what’s actually sellable on the floor and what a store will need.
Inventory Management
Items with recent data anomalies (like a UPC change) trigger a count request, so the item shows up on the to-do list when an associate starts inventory counts in Afresh.
Production Planning
Inventory positions feed into prep and bake schedules so teams plan against what’s actually available and likely to sell.
Period-End Inventory
Afresh’s understanding of inventory position updates daily, so at period close, Afresh can pre-populate estimates for every item. Teams review and adjust the items that need attention instead of starting every count from zero.
DC Buying
At the DC level, the same probabilistic model keeps inventory positions calibrated across the supply chain. Buyers size weekly purchase orders and evaluate opportunity buys against what stores can actually move and what the DC can actually fulfill.