The gallon of milk was not suddenly gone. Its on-hand count drifted down over nine days, the Tuesday order passed with the item sitting just above the reorder line, the weekend sold the rest, and the first report of an empty shelf came from a customer who walked out without it — and without the rest of the basket they came in to fill.
Stockouts are drift, not accidents
A stockout is rarely a surprise event. It is a slow mismatch between three ordinary numbers: how fast an item is selling this week, how much is on hand, and how long the next order takes to arrive. Velocity moves quietly — a hot week doubles water sales, a nearby closure sends new traffic your way, a school holiday changes what families buy. Order cadence hides the drift. If the supplier delivers Tuesdays and Fridays, an item that dips below its comfort level on Wednesday morning is not late yet, but it is already committed to running out before Friday's truck unless someone notices in time. Nobody notices across four thousand SKUs.
The arithmetic worth automating
The core figure is days of cover: units on hand divided by average daily units sold over the trailing weeks. When an item's days of cover fall below the days until its next realistic delivery — order day plus supplier lead time — it is going to run out, and that is knowable before the shelf shows it. A person can hold this arithmetic in their head for the twenty items they worry about. A nightly job that reads the point-of-sale data can hold it for every item in the store, recompute it after each day's sales, and rank what needs attention tomorrow morning.
The count that lies
The harder problem is phantom inventory: the system says six on hand while the shelf and the back room say zero. It comes from ordinary causes — shrink, a cashier scanning one flavor four times for a customer's four different flavors, receiving keyed against the wrong item, breakage nobody recorded. Reorder logic that trusts the count will calmly never reorder, because six is plenty. The detectable signature is an item showing positive on-hand and zero sales for days while the rest of its category sells normally. That pattern cannot prove the shelf is empty, but it produces exactly the short list worth walking with a clipboard, which beats hearing about it from a customer.
What the signal looks like
The output is a morning digest, not an alarm feed. Items ranked by how soon they run out relative to their next order window, with suggested quantities rounded to real case packs — nobody orders three units from a twelve-count case — and sorted into order today, walk the shelf and verify, or keep watching. It arrives before the ordering window closes, in an inbox or printed at the counter, because a signal that arrives after the order is placed is trivia.
Why it drafts orders and never places them
The system proposes and a person approves. That is a design decision, not a limitation. The math cannot see next week's ad, the seasonal reset, a substitution the rep mentioned in passing, or the cash position that says order light this cycle. A wrongly worded automatic text costs an apology. A wrongly placed automatic purchase order fills the back room with a mistake you pay invoices on. Every action that spends money passes through a human gate, which is the same principle we apply to anything that acts on a business's behalf.
What it cannot fix
The signals are only as honest as the counts underneath them. If cycle counts do not happen, the on-hand numbers rot and every calculation degrades with them. Direct-store-delivery vendors — bread, chips, soda — stock their own sections, so a gap there is a vendor conversation, not a purchase order. New items have no sales history to compute from. Promotions poison the trailing average: a velocity window that includes the ad week will over-order the quiet week after it. None of this is fatal. All of it means the digest is an instrument, and instruments need occasional calibration against the physical shelf.
How you would measure it
Keep a simple tally: stockouts discovered by a customer or a clerk versus stockouts flagged by the digest first, and for your top items, the number of days a shelf spot sat empty. Skip the lost-sales estimates vendors like to hand out. Empty-shelf days on the items people drive to you for is a measurement you can trust, because you took it yourself.
Where to start
Pick the twenty items you cannot afford to be out of. For two weeks, note their on-hand count each evening next to the delivery schedule. If that page shows items that would have run out without a mid-week scramble, you have already proven the signal is worth automating — with data from your own store.