Cycle counting: measure the first count before correcting stock

Cycle counting is the physical counting of selected inventory on a recurring schedule. It checks whether stock records match what is actually in the warehouse. For a manufacturing buyer, that matters before an MRP suggestion turns into a purchase order or a recorded balance is promised to a build.

A cycle-count report can show 95% within tolerance while only 80% of first counts match exactly. After adjustments, every checked balance may agree. Those figures answer different questions. Keep the original result, the validated count and the approved correction separate before deciding that inventory accuracy has improved.

What exactly are you counting?

Define the record before setting a target. One part in three bins can mean three records, not one. Include the site, location, part and unit of measure, plus revision, lot, serial number, ownership or stock status where those distinguish your inventory. Ten pieces in quarantine are physically present but may not be available for production.

State whether the count covers one item-location record or an entire location and all its contents. Check the floor against the list as well as the list against the floor: an unexpected item or stock in a supposedly empty bin must not disappear from the exercise. Document how new records found during the count enter the scope and report them separately if they were outside the selected list.

This is physical counting, not cycle stock, the inventory consumed between replenishments. It is also an operational control plan, not a declaration that statutory inventory or financial-statement requirements are met. Have the responsible finance or audit owner confirm the procedure your organization needs.

How do you build a cycle-count schedule?

Start with the ABC value classes, then add movement frequency, discrepancy history and production criticality. A cheap seal with no substitute can merit more attention than its C class suggests. Keep planned counts and counts triggered by a shortage, an empty bin or an unusual adjustment identifiable in the report.

Here is an illustrative plan for 1,000 item-location records at one site. The frequencies are planning choices, not a standard or a legal minimum.

With 240 counting days, that is 12.5 planned count events a day on average. Build a calendar that completes all 3,000 due events and reserve time for recounts, research and approvals. Measure the time per count in your own bins; a sealed box and a mixed bin do not take the same effort. Do not divide 1,000 by 240 and overlook repeated counts.

Track both completed events and unique records covered. Counting the same 100 records repeatedly is useful control work, but it does not establish the accuracy of the other 900. A risk-selected count program is not automatically a representative statistical sample of the whole site.

What does “95% accurate” mean in a cycle count?

Define the calculation and tolerance before counting. For an exact first-pass match rate:

First-pass exact-match rate = first counts that exactly match the original system quantity ÷ completed first counts × 100%.

Compare quantities in the same unit and at the same cutoff time. Keep missing or skipped counts out of the completed denominator but show them as incomplete work. Do not count a recount as a new successful first count.

Consider 100 distinct, risk-selected records with positive system quantities. For this illustrative example, assume an independent physical check establishes reliable reference quantities for all 100 records. Initially, 80 first counts match exactly, 15 differ but fall within the documented quantity tolerance, and 5 exceed it. Investigation and recounts then show that 5 of the 20 initial mismatches were counting errors: the original system quantities were correct. The remaining 15 records are validated as wrong and receive approved corrections.

The 80% figure measures first-pass agreement, not proven ERP-record accuracy. A first count can be wrong. The 85% figure evaluates the original records against the validated outcome in this example. The final 100% shows reconciliation of these records at that time; it does not prove the process stopped creating errors. The comparison still depends on reliable physical validation: making a ledger match entered counts is not proof those counts were correct. None of these risk-selected results establishes a site-wide accuracy percentage.

Oracle's inventory reporting documentation uses the first count and excludes recounts in its hit/miss analysis. That report also applies its configured tolerances, so it is not automatically an exact-match measure. Label your own report with the definition, scope, period and selection method.

Are tolerance and adjustment approval the same thing?

No. A measurement allowance, a reporting tolerance and permission to post an adjustment serve different purposes. A scale's uncertainty may justify a documented allowance for a bulk material. A quantity or value approval threshold decides when another person must review a correction. Neither makes an unexplained difference disappear.

Oracle's tolerance guidance distinguishes these controls. Treat it as a documented implementation example, not a claim that every ERP behaves identically. Record the actual settings your warehouse uses.

Choose tolerances for the material and control objective with the responsible inventory and finance owners. Do not import a generic “A is 0%, B is 2%, C is 5%” rule. Handle zero system stock explicitly: a percentage variance divides by zero there, but finding five unrecorded units still needs investigation. Keep wrong location, revision, lot or status visible even when the total quantity agrees.

How do you count without moving the comparison underneath it?

Use a controlled comparison time and a blind first count, with the expected quantity hidden from the counter. The GAO's final 2002 inventory-count guide explains blind counting, movement cutoff and defined error measures. Its control principles are useful here; its historical examples are not current performance benchmarks.

Do not assume that opening a count task freezes stock. Microsoft's current cycle-counting documentation explicitly says open count work does not itself block on-hand inventory in that workflow.

For the counted scope, either pause movements under an agreed procedure or capture and reconcile every receipt, issue and transfer against the count time. Record the system snapshot time, physical count time and relevant transaction references. If you cannot reconstruct the movements, recount under controlled conditions. Otherwise a transfer during counting can look like missing material.

Retain the original entry when a second counter checks a discrepancy. Record the validated result, reason, approver and adjustment reference separately. Recounting until a number matches, while discarding the earlier attempts, hides both counting problems and record problems.

What should a buyer do with a confirmed discrepancy?

First establish usable supply. If the system holds 100 units, the validated count is 90 and 95 are required for a build, the immediate physical gap is 5 units, assuming all 90 are released and unallocated. Check other locations, allocations, quality holds and open receipts before placing another PO. An unposted receipt may explain the balance without requiring a new purchase.

Once the authorized correction is recorded, rerun or review the affected MRP requirements. Confirm the dates and quantities on existing POs through purchase order tracking, then decide whether to transfer stock, reschedule the build or expedite supply.

Assign the underlying cause to an owner: an unposted receipt, wrong-bin issue, pack conversion, scrap posting or counting error needs a different fix. Review whether the same record fails again. Report gross adjustment value as well as the signed net total: a +$500 adjustment and a −$500 adjustment net to zero but represent $1,000 of gross adjustments.

Scout Agent can contact suppliers by email and phone to confirm orders and expedite late or short supply within your team's rules. Physical counting, validation and approval of stock corrections remain with your inventory process. Supplier follow-up starts after you have established what is actually missing.

Questions teams ask

Does cycle counting replace an annual inventory count?

Do not infer that from the method's name. Cycle counting is an operational control. Finance or audit owners must confirm which inventory procedure and evidence satisfy the requirements applicable to your organization.

What is a good cycle-count accuracy target?

Set the target only after defining the record, error, tolerance, selection method and time window. An exact-match rate, a within-tolerance rate and a corrected-balance rate are not interchangeable. Keep first-pass results and validated record errors separately visible.

Should a successful recount erase the first mismatch?

No. Preserve both attempts and the validated outcome. The recount may reveal a counting error, a transaction-timing issue or a genuine record error. Keeping the history lets you improve the right process without inflating first-pass performance.

Can ABC counts estimate accuracy across the whole warehouse?

Not automatically. A program that counts high-risk records more often overrepresents them, and repeated counts are not distinct records. Report the counted scope. A site-wide estimate needs a suitable sampling design and explicit treatment of coverage and uncertainty.