"Inventory is accurate" is three separate claims wearing one sentence.
They fail for different reasons, they are fixed by different controls, and treating them as one problem is why inventory work so often feels like it is not converging. You tighten the counts and the margin is still wrong. You fix the costing and the balance sheet still does not tie.
The three assertions:
Quantity. The system count matches the physical count. Valuation. The unit cost on the books is the right cost. Assignment. That cost lands in the right period, location, and channel.
Take them in order, because quantity is the foundation and the other two are built on it.
Quantity, and why cycle counts beat an annual count
An annual physical count is a point in time assertion. It tells you what was true on one day and says nothing about the other 364.
Worse, it tells you at the least useful moment. If the count reveals a systematic receiving problem, that problem has been running all year and every monthly margin figure you reported was affected by it. You cannot go back and fix eleven months of decisions.
Cycle counting replaces one large assertion with a continuous stream of small ones. The design questions are:
What to count and how often. ABC stratification is the standard approach and it works: high value or high movement items counted frequently, slow low value items counted rarely. The goal is to spend counting effort where being wrong is expensive.
Who counts. Not the person responsible for the stock. Independence is the whole point of a count, and a count performed by the person whose accuracy is being measured is a self assessment.
What you do with the result. This is where most programs fall down. If the output is "we counted and adjusted," you have a correction process, not a control. Accuracy needs to be a tracked metric, by location and by category, so you can see where it is degrading while you can still act.
One detail matters more than it looks: reason code your adjustments honestly. If every variance is coded "adjustment," you have a number with no information in it. Receiving error, unrecorded movement, damage, and miscount point at four completely different fixes, and only the reason codes tell you which one you have.
A useful diagnostic: compare SKU level accuracy against unit level accuracy. If unit accuracy is high but SKU accuracy is low, you have many small variances, which points at process and measurement discipline. If SKU accuracy is high but unit accuracy is low, you have a few large ones, which points somewhere more serious.
Valuation, and the quiet drift of standard cost
Standard cost exists to give you a stable basis for margin analysis and to make variances visible.
Both of those break down in the same way, and it is easy to miss. A standard that has drifted far from actual has not stopped working. It has started lying quietly. The variances still calculate correctly, they are just now absorbing a structural gap rather than flagging a problem, and the accumulation hides the drift.
The practical disciplines:
Set a cadence for revisiting standards and keep it, including in periods where nothing felt like it changed. Cost drift is gradual and nobody notices the day it becomes material.
Watch purchase price variance and usage variance separately. They tell you different things. Purchase price variance is a sourcing signal. Usage variance is a production or measurement signal. Blending them loses the diagnostic value of both.
Investigate variances that are consistently one directional. Random variance around zero is normal. Variance that is always unfavorable for the same item is a standard that needs updating, not a performance problem.
The other half of valuation: what the goods are actually worth
Getting the cost right is only correct if the goods are still worth that cost.
Inventory is carried at cost, but only until cost stops being recoverable. Slow moving stock, superseded product, and anything damaged or returned in unknown condition all raise the same question: would you realize this amount if you sold it today, net of what it costs to sell?
The practical control is aging inventory the way you age receivables. Quantity on hand against recent movement, by category, so that stock which has not moved in two quarters is visible as a question rather than sitting quietly at full cost.
Noticing it early gives you options: a promotion, a channel change, a bundled sale. Noticing it at year end, when the auditor asks, gives you a write-down and a conversation about why nobody flagged it.
Assignment
The third assertion is the one people forget is an assertion at all.
The right cost in the right period, at the right location, allocated to the right channel. In a business selling the same unit through several channels, this is where inventory accounting meets margin reporting, and where a correct quantity and a correct unit cost can still produce a wrong answer.
The most common failure is timing. Goods received at the end of a period but invoiced in the next one, or shipped before period end and billed after. Both distort inventory and margin simultaneously, and both are caught by the same control: a cutoff review that compares receiving and shipping activity around the period boundary against what was recorded.
Inventory you do not physically hold
Third party fulfillment, in transit, consignment. Ownership does not follow possession.
Goods at a 3PL are yours. Goods shipped FOB origin stop being yours when they leave your dock. Consignment stock at a customer remains yours, along with its obsolescence risk, until it sells.
The hard part is counting what you cannot walk up to. Either you reconcile to the third party's reported counts on a defined schedule and investigate differences, or you accept that a meaningful portion of the balance sheet is unverified. There is no third option, and the second one is a finding waiting to happen.
The reconciliations that catch problems
Three, and they catch different things:
Sub-ledger to general ledger. Catches posting and interface errors. Should be clean every period, and a persistent difference is a system problem rather than a counting problem.
System to physical. The cycle count program above.
Receiving to invoice. Catches the quantity and price discrepancies that would otherwise flow into inventory value undetected, which is also the three way match that protects accounts payable. One control, two benefits.
The short version
Count continuously rather than annually, and track accuracy as a metric rather than a pass or fail event. Reason code honestly or the counting tells you nothing. Revisit standards on a cadence. Age your inventory so recoverability is a question you ask early. Reconcile all three ways.
If your margin analysis feels unreliable and you cannot say why, work down that list in order. It is almost always one of them, and it is usually the counts.