Inventory accuracy: how it is measured and raised.

How inventory accuracy is measured

Inventory accuracy is usually quoted as a single percentage, meaningless without knowing what sits in the denominator. The version worth trusting is location level: the share of racking positions where the item recorded in the system matches what physically sits there, including positions the system expects to be empty. A count that only checks positions the system says are full will never catch a phantom pallet, because it never looks anywhere the system did not already point it, which flatters the figure it produces.

A single pass only measures a moment. What most businesses actually want to know is whether accuracy holds between counts, a different question with a different measurement approach, covered under methods below. Sentispec Inventory's own method, including what a camera based count does and does not verify, is set out in detail on how accuracy is measured.

Why records drift

Stock records do not drift because people are careless. They drift because every touch of a pallet is a chance for the system and the shelf to disagree, and a warehouse generates thousands of touches a day.

• Receiving errors. A pallet booked in against the wrong SKU, quantity or location starts wrong and stays wrong until someone happens to notice.
• Put away shortcuts. A driver puts a pallet where there is space rather than where the system was told, especially under pressure, and the transaction never catches up to the shelf.
• Picking without confirmation. A partial pick, a substitution, or a pick recorded before it physically happens all leave a gap between the transaction log and reality.
• Damage and write offs handled outside the system. Stock removed informally, without a matching adjustment, quietly inflates what the system believes is still on the shelf.
• Returns and reverse logistics. Stock re-entering the building through a different process than it left by is one of the most common sources of a record nobody can reconcile later.

Each of these is small on its own. Across a year, on a site with tens of thousands of positions, they compound into exactly the kind of gap a stocktake exists to close. The specific mechanisms are covered in more depth, with what each looks like in a discrepancy report, in causes of stock discrepancies.

Annual counts, cycle counting, continuous verification

The three common approaches to keeping accuracy in check are not interchangeable, and the right one depends on how much a discrepancy costs you while it goes unnoticed.

The annual count is the oldest method and the one auditors are most used to seeing, but it answers a question about a single day, not the year around it. Cycle counting improves on that by rationing attention toward the stock that matters most, typically using ABC classification, and is a sound compromise as long as counting stays expensive. Continuous verification is what becomes possible once it does not: when a meaningful count fits inside a break rather than a shift, there is little reason to leave any class of stock uncounted for months. The frequency question is worked through in detail, including a worked frequency table, in how often should you cycle count.

Six ways to raise accuracy, and their trade offs

The six routes are listed in short at the top of this page. What follows is what each one actually asks of you once you commit to it.

• 1. Tighter receiving and put away discipline. The cost is supervisory attention, and it lands on the two moments a warehouse most wants to speed up. It is also the only route here that asks operators to be more careful rather than changing what has to be done at all.
• 2. Barcode or RFID scanning at every touch. It timestamps every movement, which is worth having on its own. It also needs infrastructure, tags and reader discipline throughout the building before any of that starts. See RFID against camera based counting.
• 3. More frequent manual cycle counts. Straightforward to start, and it directly reduces how long a discrepancy can hide. Because the cost is linear, most sites ration it by stock class rather than counting everything more often, which leaves the C items drifting.
• 4. Master data and location cleanup. Duplicate location codes, badly labelled bays and stale SKU records cause discrepancies that have nothing to do with the physical count. Drift resumes once the data is clean, so it is worth doing early and worth repeating.
• 5. Camera based continuous counting, such as Sentispec Inventory. The throughput is the whole mechanism: nothing about counting more often is difficult once it stops costing a shift, and a 20 minute run covers 500 to 1,000 pallets. The limit is the driver, which is a real cost against a method that counts unsupervised.
• 6. Autonomous robots or drones. The genuine advantage is a site where labour is scarce or aisles run unmanned. The facility has to tolerate a battery powered vehicle in the racking, which a cold store does not always do well. See inventory drones and autonomous ground robots.

What 99%+ looks like day to day

A percentage hides the size of the actual problem sitting behind it. On a 30,000 location warehouse, the difference between 95% and 99% accuracy is the difference between roughly 1,500 wrong positions and roughly 300, and the difference between 99% and 99.9% is the difference between 300 and 30. That last gap is the one that decides whether discrepancies are a managed queue one person clears in a morning, or a routine part of how pickers expect their day to go. The full arithmetic, and what each accuracy band feels like operationally, is set out on how accuracy is measured.

Reaching 99%+ once is a project. Holding it is a habit, which is why continuous verification tends to outperform an annual reset over a full year. A site counted once a year is only briefly at its best figure, immediately after the count, and spends the rest of the year drifting away from it.

Building the internal business case

Whichever of the six routes above you choose, the business case rests on the same three numbers: what the current gap costs you, what closing it costs, and how fast the difference pays back. Start with what a stocktake and the searches around it currently cost, not what you assume they cost. That figure is usually larger than the labour line alone once lost pallet search, write offs and any production downtime tied to stock errors are added in, and it is broken down in full on the cost of manual stocktaking.

From there, the payback arithmetic, what finance will challenge in it, and what to put in front of a CFO versus an operations director are covered step by step in building the business case for stocktaking automation. That guide also lists the questions worth asking any vendor pitching a fix, Sentispec included.

Four practical guides cover running a count, counting frequency, the causes of drift and the business case, all indexed on the guides hub. Terms used above are defined in the glossary.