The board pack shows 97% inventory accuracy across the network, and the operations director presents it as a win. Three weeks later, a customer order is short on a critical sub-assembly because the finished goods bay at the secondary plant has been miscounting for months. Nobody saw it coming, because nobody was looking at the right number. The 97% was a weighted average that buried a site running closer to 82%. The good locations carried the bad ones in the math, and the math told leadership everything was fine.
This is the most common pattern in inventory measurement. A single rolled-up accuracy figure is comforting and useless. It tells you nothing about which dock is bleeding stock, which warehouse is over-counting receipts, or which production floor is silently absorbing variance into the cost of goods sold. Real inventory accuracy calculation has to start at the location level, and it has to be honest about what counts as a discrepancy. The honest version is rarely the version on the executive dashboard, because the honest version usually looks worse.
This post explains how to compute per-location accuracy properly, why adjustment events are the most reliable early warning signal, and what benchmarks to expect once you start measuring at the right level of granularity.
The Inventory Accuracy Formula That Actually Means Something
The textbook inventory accuracy formula is straightforward. Take the number of stock keeping units that match the system count exactly, divide by the total number of stock keeping units counted, and multiply by 100. The result is a percentage. The problem is not the formula. The problem is what gets fed into it and how the result gets interpreted.
A site that counts 500 unique items and finds that 485 match the system reports 97% accuracy. That sounds healthy. It is not. Fifteen items off in a manufacturing context can mean fifteen production stoppages, fifteen emergency expedites, or fifteen unhappy customers. The denominator hides the operational severity of the numerator. Worse, the formula treats a one-unit discrepancy on a slow-moving fastener as identical to a fifty-unit discrepancy on the constraint material for next week's largest production run.
A more useful version of the inventory accuracy calculation weights the result by movement volume or by dollar value. A weighted formula divides the absolute count-to-system delta by the total quantity counted, then subtracts that ratio from one. So if you count 10,000 units across an item set and the absolute delta sums to 300 units, your weighted accuracy is 97%. The same 97%, but now it represents the share of physical inventory that matches, not the share of items. That distinction matters when you are trying to size the financial exposure rather than just count failures.
For a location-level view, run this calculation per site, per item class, and per cycle count event. The number you should be reporting upward is the lowest performer in the network, not the average. A network is only as accurate as its weakest location, because that is where the next surprise will come from.
Why Org-Wide Accuracy Hides Bad Sites
Most ERPs and inventory systems aggregate accuracy by default. They show a single number on the home screen, often with a green arrow if it ticked up since last quarter. The aggregation is mathematically defensible and operationally dangerous. A network with five warehouses running at 99% and one warehouse running at 80% will report somewhere around 96% on a weighted basis, and the 80% site will be invisible.
The hiding effect is worse when locations have different sizes. A small warehouse with 200 SKUs running poorly contributes very little to a network total dominated by a 5,000 SKU central distribution center. The smaller site can be in genuinely critical condition, and the rolled-up number will not budge. By the time it does, the operational damage is already done.
Per-location stock records solve the measurement problem at the data layer. When every location has its own stock record per item, accuracy can be calculated independently for each site. This is the foundation that makes location-level accuracy possible. Without it, you are always reverse-engineering site numbers from a global pool, and the math gets ugly fast. As covered in the analysis of why spreadsheet inventory fails at scale, flat files cannot maintain this separation cleanly because they have no native concept of a location dimension that links to a movement history.
The reporting discipline that follows is simple. Show the location with the lowest accuracy first. Show the count-to-system delta in absolute units and in dollars. Show the trend, not just the snapshot. If a site has dropped from 96% to 91% over three count cycles, the snapshot is less important than the slope. The slope tells you whether a process is breaking down.
Adjustment Events as the Canary
Cycle counts happen on a schedule. Adjustment events happen continuously. If you wait for the next quarterly count to discover a problem, you have already paid for three months of bad data. The adjustment ratio, defined as the volume of adjustment movements relative to all movements at a location over a window, is the closest thing operations has to a real-time accuracy gauge.
Every time someone records a cycle count discrepancy, a damage write-off, a found-in-the-back recovery, or a manual correction, an adjustment movement gets created. In a healthy operation, these events should be rare and small. In an unhealthy operation, they spike. The spike is usually the first visible symptom of a process problem, and it shows up weeks or months before the next scheduled count would surface it.
Track the adjustment ratio per location, weekly. A site whose ratio doubles in a month is telling you something. Maybe a new operator was onboarded without proper training. Maybe a scanner is misreading lot numbers. Maybe a transfer process is dropping units between dispatch and receipt. The cause varies, but the adjustment ratio is the signal that demands investigation.
Immutable adjustment events make this analysis possible because they cannot be edited or deleted. As detailed in the discussion of why every movement matters in an immutable audit ledger, every change to stock generates a permanent record with timestamp, actor, quantity, and movement type. When you query for adjustments at a specific location over a date range, you get the complete picture, not a sanitized version that has been cleaned up to look better in retrospect. This is the difference between an accuracy metric you can trust and one that drifts based on who edited the underlying data last.
Setting Honest Location Accuracy Benchmarks
A reasonable location accuracy benchmark depends on what you are storing, how you are counting, and how mature your processes are. A finished goods warehouse with serialized product, scanner-driven receipts, and weekly cycle counts should run above 99%. A raw material store with bulk components, manual issue logging, and quarterly counts will rarely exceed 95% in honest measurement. A factory floor with active consumption, partial returns, and informal staging areas often runs in the 85% to 92% range, even with disciplined operators, because the environment generates more legitimate variance.
The mistake is to apply a single network-wide target to all location types. A 98% target imposed on a factory floor will either be missed every quarter or hit through creative accounting. Neither outcome is useful. Set targets per location type, calibrated to what is achievable with the actual processes in place. Then drive improvement against the location-specific target, not against an imaginary uniform standard.
Per-location alerts make this calibration enforceable. When a stock record drifts below its threshold or when an adjustment volume spikes above a configured limit, the alert fires for that location only. The plant manager responsible for the site sees the alert. The corporate director sees a roll-up of which sites are flagging. Nobody is forced to wade through irrelevant notifications, and nobody is shielded from problems that are actually theirs to solve.
Cascading location health extends this principle upward. A location is healthy if its stock records are healthy. An organization is healthy if its locations are healthy. The cascade preserves the ability to drill down from any aggregate view to the specific record causing the problem, which is the only way leadership can act on accuracy data without becoming a bottleneck themselves.
Closing the Loop on Location Accuracy
The version of inventory accuracy that drives operational improvement looks nothing like the version that goes in a quarterly presentation. The presentation version is one number, smoothed across the network, designed to communicate stability. The operational version is a table, sorted by worst location first, weighted by units and dollars, refreshed against an immutable adjustment history, and benchmarked against location-type-specific targets.
If you are reporting one number upward, you are managing one number, and you are missing the locations that need attention. Switching to per-location measurement is uncomfortable at first because the picture gets worse before it gets better. Sites that looked fine in the aggregate suddenly show as 87%, and the temptation is to argue with the methodology rather than fix the underlying processes. The argument is a distraction. The methodology is the point. When every location has to report its own honest accuracy and its own adjustment ratio, the conversation shifts from defending the network number to investigating the location number, and that is where the actual operational work gets done. Visit https://falorb.com to see how location-scoped accuracy reporting is implemented in practice.
FalOrb helps manufacturers measure inventory accuracy honestly at every location through location-level stock records, immutable adjustment events, and cascading health visibility. Book a 30-minute walkthrough or email us at [email protected] to see how it applies to your operation.