A perpetual inventory system promises a stock figure that is always current. That promise is real, and it is why most growing operations move to one. For how the method works, start with the perpetual inventory topic. The live number is not free, though, and the costs are easy to underestimate before you have lived with them for a year.
Key takeaways
- The core weakness is drift. A perpetual record is only as accurate as the capture discipline behind it, and the research says that discipline is worse than most operators assume: a study of nearly 370,000 records across 37 stores found 65% of inventory records inaccurate, with an average absolute deviation of about five units, roughly 35% of the units actually on the shelf.
- Perpetual does not end physical counting. It converts an annual stop-the-line count into ongoing cycle counts, which is usually better and is still real labour.
- The expensive failure is not the wrong number. It is automation acting on the wrong number, because reorder points and replenishment rules execute without pausing to doubt the input.
- Costing gets harder, not easier. Perpetual FIFO or weighted average requires a cost calculation at every transaction rather than one at period end.
- It earns its cost when being wrong is expensive: tight margins, fast turns, automated reordering, or promises made to customers against available stock.
The record drifts from reality, and the drift is measurable
The recorded quantity stays accurate only if every movement is captured. In practice some are not. A unit walks out unbooked. A return sits in a corner for a week. A scan reads the wrong SKU. None of these throw an error, which is the whole problem. The book quantity diverges quietly from the shelf and you keep trusting it until a count proves you should not have.
The scale of that gap is not a matter of opinion. DeHoratius and Raman’s analysis in Management Science examined close to 370,000 inventory records across 37 stores of a single retailer and found 65% of them inaccurate. The average absolute deviation between system and actual was almost five units per SKU, about 35% of the average quantity on the shelf. That is a retail environment rather than a warehouse, and a disciplined operation will do better, but the direction is the point: perpetual records drift by default and stay accurate only where something actively holds them in place.
Where the drift comes from
Six sources account for most of it, and they are worth naming because each has a different fix.
- Unbooked movements. Stock that moves without a transaction: samples, internal use, a rush pick done by hand during a busy hour.
- Receiving errors. Quantity accepted on trust, pack sizes assumed rather than counted, a pallet booked as received before it is verified.
- Unit of measure and pack size. The single most common systematic error. A case booked as an each, or an each booked as a case, drifts the record by a multiple rather than by one.
- Mis-scans and substitutions. Similar barcodes, a damaged label, a picker scanning the nearest identical carton.
- Unrecorded returns. Goods back in the building and not yet back in the system, which shows as available stock that no picker can find.
- Shrink. Theft, damage, and administrative error. The last industry-wide benchmark from the National Retail Federation put shrink at 1.6% of sales in FY2022, about $112.1 billion, up from 1.4% the prior year. Worth knowing when you cite that figure: the NRF discontinued the annual shrink survey in October 2024, saying a broad study of retail shrink was no longer sufficient, so there is no newer number of the same kind to compare against.
Our deeper treatment of the causes and the detection side is in retail shrinkage and phantom inventory detection on transaction streams.
It demands operating discipline you cannot buy
Perpetual accuracy is an operations habit, not a software feature. Receiving has to book stock before it moves. Picking has to confirm what left. Adjustments have to be entered the same day, not on Friday. A team that treats these as optional will run a perpetual system that is perpetually wrong, and the software will report its wrong number with total confidence.
This is why the cost of a perpetual system is rarely the licence. It is the training, the process design, and the fact that discipline decays with staff turnover and has to be rebuilt each time.
It does not remove counting
The most common misconception is that real-time tracking ends physical counts. It does not. It changes them from an annual stop-the-line event into ongoing cycle counts that reconcile book to shelf. That is usually a better way to count, and it is still counting.
Two consequences follow. First, cycle counting is recurring labour that has to be staffed and scheduled, not a project. See how cycle counting works and the physical count versus cycle count comparison for the mechanics. Second, the sampling design matters more than most programmes admit: counting a convenient subset and extrapolating is not the same as knowing your accuracy, an argument we make in full in cycle count sampling is statistically indefensible. Use ABC analysis to decide what gets counted often, because counting everything at the same frequency wastes the budget on items where being wrong is cheap.
Automation amplifies whatever the record says
This is the disadvantage that actually costs money, and it is a consequence of the system working as designed rather than failing.
A periodic system is checked by a human before it drives a decision. A perpetual system is wired into replenishment. When the record says 40 and the shelf holds 12, the reorder logic does not hesitate: it sees a quantity above the reorder point and orders nothing, and the stockout arrives with every setting correct. The reverse case is worse in cash terms, because a record that overstates nothing on hand triggers purchase orders for stock already in the building.
The formulas downstream inherit the error and have no way to detect it. That is true of the reorder point and safety stock calculation and of economic order quantity alike. Both assume the on-hand figure is a fact. Automating replenishment on a drifting record does not create the drift, it just converts it into purchase orders faster than a human would.
Costing gets harder, not easier
A periodic system computes cost of goods sold once at period end from opening stock, purchases, and a closing count. A perpetual system computes a cost at every transaction, which means the costing method has to run continuously: FIFO layers maintained per receipt, or a weighted average recalculated on each purchase.
The accounting is more useful, because margin is visible during the period rather than after it. It is also more places for an error to enter, and errors in cost layers are harder to spot than errors in quantity, since nothing on the shelf contradicts them. The perpetual and periodic comparison walks through both mechanics side by side.
So when is it worth it?
When the cost of being wrong is high: tight margins, fast turns, automated reordering, or customer promises made against available stock. If volume is low, the catalogue is small, and a stockout costs an apology rather than an account, a periodic system may cost less to run and lose you very little.
The decision is about the economics of your operation, not about which system is more modern. A perpetual system run without capture discipline is strictly worse than a periodic one, because it produces a confident number nobody has any reason to trust, and confident wrong numbers travel further than admitted ignorance.
The verdict
Perpetual inventory is the right default for most operations above a certain size, and its disadvantages are not arguments against it. They are the running costs: capture discipline, cycle counting, reconciliation labour, and a costing method that has to be maintained transaction by transaction. Price those honestly before the implementation, because they arrive whether or not they were budgeted.
If you want a second set of hands on inventory accuracy and lot-sizing policy across the catalogue, AvanSaber’s inventory advisory practice does this work case by case.