Blog
Fresh inventory needed its own standard, so we built one
Published
Connect with us
Before Afresh, grocery inventory systems didn’t account for shrink because they simply couldn’t see it. The gap meant on-hand estimates in fresh couldn’t be trusted, which led to poor planning decisions. We were the first to build a model that worked out what happened inside that gap to calculate fresh inventory more accurately, and our work in inventory estimation recently earned a patent.
Before Afresh came on the market, grocery technology systems relied on perpetual inventory (PI) for on-hand estimates. PI works for cans of soup because most movements are traceable: yesterday’s on-hand, plus what came off the truck, minus what rang through the register.
In fresh departments, however, many moves go unrecorded: browning bananas get culled and rarely scanned out, and trimmings from the seafood counter get tossed. Without clear records, PI drifts fast, and even the best forecasting system will recommend inaccurate orders when it starts from the wrong inventory number.
If it were easy to fix, someone would have done it before Afresh began nearly a decade ago. The issue isn’t only that shrink hardly gets logged. It’s also incredibly complex to measure and predict. Item perishability differs depending on the vendor, travel, season, and handling. It’s not just a missing number in an equation, but rather a whole range of possible outcomes.
Since it seemed impossible to calculate, retail inventory systems simply left it out.
When Afresh began, our north star was building technology for grocery’s idiosyncrasies instead of ignoring them. In an industry with margins so narrow, getting every plan right matters. Our continued innovation in this space recently earned a patent, so we thought it was a good chance to look back at how we solved this decades-old problem.
Mapping what can’t be seen
A store’s inventory system sees a handful of recorded events: deliveries, sales, counts, and whatever discards get scanned. Everything in between is invisible: product gets bruised and tossed, mis-scanned, moved to the backroom, or cut into another item.
We believed that an accurate inventory estimate had to consider shrink and that shrink estimates could be worked out based on other signals, including our perishability modeling.
Our patented method estimates how quickly each item spoils at different times of year, how likely items are to go unsold, get stolen, or be mis-scanned, and how they transform. That means our Intelligent Inventory accounts for the six clamshells of strawberries pulled on the morning cull walk, the bag of limes self-scanned as lemons, and the whole watermelons cut down into fruit cups for the grab-and-go case.
A game changer for accuracy
The inferred estimate went into the inventory calculation, and the output was an estimate 10% to 40% more accurate than a perpetual inventory estimate on perishable items.
Because the inventory model built its own picture of what was on hand, orders, production plans, and ending inventory counts became more accurate. It also let us significantly reduce the number of inventory counts store teams took each day. Counts then became evidence the estimate was weighed against, not the starting point for it, so a count with an extra zero in it didn’t skew the whole picture.
Finding and solving gaps that are unique to grocery is how we approach every problem at Afresh. With this model, grocers could finally get inventory numbers they could trust, and forecasting systems could generate more accurate recommendations that led to a 3% increase in sales and an 80% reduction in stockouts.
For readers who want the more technical details on our inventory estimation, read our engineering blog on the topic.
Congratulations to our inventors, Clement Micol and Aaron Stern.



