Warehouse Operations Guide

Warehouse Cycle Counting for 3PLs: A Practical SOP for Inventory Accuracy

Warehouse cycle counting checks a defined slice of physical inventory against records at an agreed time. Start with the owner, location, item, tracking attributes and unit. Control or reconstruct movements, count, investigate differences, then decide what can resume. A variance is a question to resolve before it becomes a quantity adjustment or a shipping decision.

Consider a hypothetical afternoon outbound problem: the system shows 40 units for a client, but staff find only 34. Six units might already be picked, moving between locations, recorded in another unit, assigned to another owner or held in another status. There may also be a real shortage. Preserve the evidence and control the affected work while checking those possibilities. Do not immediately write off six units.

This suggested SOP is for ordinary ecommerce inventory in small and midsize US 3PLs. Receiving verifies newly arrived goods; cycle counting checks stock already under warehouse management. Batch picking organizes order collection. This guide addresses ongoing inventory accuracy and discrepancy handling within the broader 3PL operating workflow. It does not replace client, financial or audit requirements for other counts.

Two staff verify a selected warehouse bin and packaged goods while another worker picks orders in a separate aisle.
Cycle count illustration. Staff check a selected bin while picking continues in another aisle.

1. Define exactly what the count will compare

An inventory cycle count checks selected stock repeatedly through the year. A full physical inventory covers the entire agreed inventory scope at a defined cutoff. Both require physical evidence and a consistent basis for comparison. Extensiv describes counts by customer and smaller groups, while ShipBob compares recurring counts with broader physical counts. Their guidance provides context; the control steps below are proposed operating procedures. Extensiv counting overview and ShipBob cycle counting overview

In a multi-client warehouse, use a count identity such as owner account + warehouse + location + item identifier. Add lot, serial, status or other required tracking attributes. State the counting unit. These are checks to agree before work starts, not instructions to redesign a database. Two clients using the same SKU text still have separate inventory ownership. A building-wide quantity total cannot establish that each owner has the right goods.

A SKU in several locations requires a clear boundary: one pick bin, all recorded bins, or a physical sweep of a defined area. Include the handling of goods found outside their recorded positions. A system-zero location can contain unrecorded stock. A blank or skipped count means not counted, not zero. Require an explicit empty confirmation after inspection, and keep inaccessible stock unresolved rather than entering a convenient number.

Agree before countingExample / boundaryWho confirms
Owner and facilityClient A; warehouse WH-1; same SKU text in Client B remains separateInventory lead
Location and itemBin P-01; item ACC-01; state whether other positions are includedLead + counter
Tracking and statusLot/serial where required; sellable, damaged and pending inspection remain distinctClient rules + inventory lead
Unit and content methodEach or carton; verify pack definition and permitted content checkItem-data owner
Time and quantity definitionBaseline timestamp, count time, on-hand definition, physical vs posting timeInventory + outbound leads
Exceptions and releaseUnexpected stock, inaccessible goods and not-counted lines stay explicitReviewer

Define on-hand and allocatable quantities using the actual system. Allocated, picked, quarantined and available fields may overlap or describe different stages; do not add them as though each were an independent pile. Ask where physically picked goods sit in the chosen report. A reservation can change availability without changing the quantity on a shelf. Correctly counted damaged stock remains damaged and may still be unavailable to orders.

Extensiv's 3PL Warehouse Manager audit FAQ distinguishes skipped lines from counted inventory and explains why missing track-bys prevent reconciliation. It also warns that audits begin from system-known stock. Treat that as a reason to add an explicit physical-location check where needed, rather than assuming a generated list discovers everything. Confirm how your own count type handles unexpected items and zero locations. Extensiv audit FAQ

For this hypothetical report comparison, save the report filters and field definitions with the task. If a screen shows 40 available units but another report shows 40 on hand, the identical numeral does not make them interchangeable baselines. Have the inventory lead resolve that definition before comparison. A different shift should be able to rebuild the scope from the saved identifiers and timestamps, including the units that were excluded and the reason.

2. Choose coverage, frequency and capacity together

Choose today's work from due planned counts, customer commitments and current exceptions. Location counts can expose misplaced goods and unexpected contents. Item counts follow a SKU across its known positions but may miss an unlisted wrong location. A risk-based schedule concentrates attention where a failure matters most. A pick shortage, negative balance or unusual return can trigger an extra count. Small random checks help test assumptions beyond the problem list.

Selection methodCan revealCan miss / control needed
LocationUnexpected or misplaced contents in the selected spaceOther locations; inspect system-zero bins explicitly
ItemA chosen SKU across its known positionsUnlisted wrong positions; add a physical sweep if needed
Risk / transaction priorityHigh-impact and recurring errorsUntouched low-priority stock; maintain due coverage
Exception-triggeredPick short, negative balance or unusual returnNot a substitute for planned coverage
Small random checksIssues beyond the selected problem listSample limitations; define the sampling frame

Build ABC priorities from value, transaction frequency, past discrepancies, unit complexity, customer commitments and order impact. A low-value fast mover can be operationally important. Repeated counting of one troublesome record does not cover an untouched zone. ShipBob discusses ABC, geographic, opportunity and control-group approaches, including their limitations. Use those as method choices, not a mandatory universal calendar. ShipBob method comparison

A workload model you can recalculate

Assume a stable planning population of 2,000 inventory records, each already defined by owner, item, location and necessary attributes. Class A has 200 records counted 12 times annually; B has 600 counted four times; C has 1,200 counted once. Over 240 counting workdays, the annual workload is 200 × 12 + 600 × 4 + 1,200 × 1 = 6,000 record-count events. The average is 25 events per day.

Hypothetical classStable recordsAnnual frequencyRecord-count events
A200122,400
B60042,400
C1,20011,200
Total2,000Varies by record6,000
Daily average over 240 workdays25 events25 × 4 minutes + 20 + 20140 person-minutes

The plan covers 2,000 distinct records through 6,000 observations. It does not cover 6,000 different records, and one WMS job can contain several planning records. Assume each event takes four minutes, including normal travel and capture once ready. Add 20 minutes for recounts and 20 for preparation and handoff each day. That is 25 × 4 + 20 + 20 = 140 person-minutes, or two hours 20 minutes daily across the assigned roles.

These are staffing assumptions, not industry benchmarks. They do not promise equal daily work or replace independent review with self-approval. New items, new locations, urgent exceptions and longer investigations need capacity beyond the model. A SKU across several locations cannot automatically use one four-minute allowance. Track due dates by record, because a manageable average can still hide overdue high-impact work.

Before scaling, time a small mix of easy bins, mixed units and discrepant records. Separate first-count time, travel, preparation, recount and investigation. Record interruptions instead of blaming the counter for all elapsed time. Use the observed range to revise the schedule and reserve, and check whether a second qualified reviewer is actually available when discrepancies arise.

Extensiv's March 2026 automation guide describes customer-enabled, item-based audit creation, with item frequencies and a warehouse daily job cap. The cap can delay due work; an automated job is not proof a count happened. Those are 3PL Warehouse Manager features, not a common function of every inventory or shipping tool. Check the account configuration before relying on scheduling automation. Extensiv automation guide

3. Control movement around the count time

A warehouse can keep operating while the selected inventory is controlled. That does not mean the same bin can be counted, picked, replenished and relocated without coordination. Choose either a short local movement pause or a supported process that reconstructs relevant movements against a clear time basis. The inventory lead and outbound lead should agree which existing tasks are affected before the counter starts.

Option A: a short local pause

List the selected locations, owners and stock attributes, then identify open picks, replenishments and transfers. Finish or explicitly park those tasks. Set the physical boundary and communicate a start time, an owner and a release signal. Use the system's supported controls plus a clear floor instruction as appropriate. Confirm whether previously generated tasks can still execute; a setting that stops new allocation may not stop an existing pick.

Once the pause is confirmed, capture the baseline and count without unlogged movements through that scope. Other verified locations can serve unaffected orders. If someone moves the selected stock, record the event and stop relying on the original comparison until it is resolved. A small team without dependable movement evidence should use a fresh controlled count, not guess what probably left the shelf.

Option B: a supported movement reconciliation

Use this only when the system behavior and complete movement records allow it. Preserve the baseline timestamp and each relevant movement's quantity, unit, source, destination, owner and physical event time. Keep the system posting time separately. Bring the expected balance and physical observation to the same point in time. Include receipts, picks, transfers and reversals that cross the chosen boundary, without counting an internal movement twice.

If a record only proves an order was released, it may not prove six units physically left this bin. If a transfer names the wrong destination, inspect the route and both locations. Missing timestamps, unclear units or duplicate postings make the reconciliation unreliable. Pause a smaller area, resolve the record gap and recount. An arbitrary historical screenshot cannot be subtracted from a physical count taken whenever staff happen to arrive.

What product documentation actually establishes

ShipHero's July 2026 management guide states that an In Progress count makes the location being counted unpickable. Verify the behavior for your count type and existing tasks; it is not evidence that every WMS locks bins, blocks replenishment or cancels work already released. When evaluating your system, test allocation, an existing pick, replenishment and a transfer in a safe setup with your administrator. Use the small-3PL WMS guide to organize that evaluation. ShipHero count management

ShipHero's recount documentation separately describes comparing the variance against an updated expected balance when activity occurs between observations. Its second-user recount setting must be enabled; it is not the default. This does not authorize uncontrolled simultaneous movement inside an active count. Document which stage allows activity and how the comparison works before adopting a similar method. ShipHero recount documentation

4. Run the count with explicit handoffs

First, the inventory lead issues a task with its scope, unit, tracking attributes, cutoff and movement method. The outbound lead checks pending work and acknowledges the affected orders or bins. Work starts when those inputs agree. If the SKU, owner or location cannot be identified, return the task for clarification. A counter should not invent the missing scope from a familiar-looking label.

Next, the lead records the baseline and confirms movement control. The counter scans or verifies the location and item, then counts the stock using the approved unit and content-check method. Scanning helps identify a label; it cannot prove a closed carton holds its printed quantity or that every unit is sellable. Record opened packs, damaged units and inaccessible contents separately so a reviewer knows what was actually observed.

Consider a blind first count, where the counter records observed quantities before seeing the expected balance. This can reduce anchoring on the system number, but it is a method choice rather than the only legitimate procedure. ShipHero's mobile guide documents a Blind Count option that hides expected quantities. Verify its behavior and report access for your users; a paper worksheet can also separate observation from comparison. ShipHero mobile count guide

Record the first result and count time before discussing the difference. Unexpected goods need their actual found location, description, owner evidence and a linked exception. Do not merge them into a similar SKU or leave them off the task because the screen lacks a row. If identity is uncertain, apply the warehouse's pending-identification process. Keep these goods from being treated as verified stock while investigation continues.

For a variance, have a different qualified person recount the same scope without simply copying the first number. Confirm the unit, tracking attributes and time basis again. Preserve both observations. If movement occurred between counts, reconcile it through the approved method; matching raw totals are not meaningful unless the comparison basis also matches. If an independent person is unavailable, document the limitation, escalate and keep the exception open under your approval policy.

The reviewer then classifies the result and assigns investigation, correction or authorized quantity action. Record completion criteria for each handoff: identified stock, valid observation, explained variance or approved resolution, and an explicit release review. A timer ending or a task being marked complete is not sufficient. Keep unresolved work assigned to a named owner, with the next check and a due time.

5. Investigate the difference and approve the right action

Work through identity and owner, location, unit conversion, lot or status, period movements, and duplicate or missing records. Use the independent recount to test the physical observation. An unexplained residual difference can then support an authorized quantity decision. Finding a mismatch and assigning responsibility are separate actions. There is no need to accuse a client, worker or carrier to record what is known.

Difference typeEvidence to checkOwner / possible action
QuantityIndependent counts, movement history, other positionsInventory reviewer; authorized quantity action if supported
LocationSource/destination scans, transfer, found positionWarehouse lead; correct the supported move
OwnerClient account, receipt, identifiers and transfer chainInventory + client contact; investigate attribution, never cross-offset
UnitPrimary/pack units, actual pack and content checkMaster-data owner; validate conversion and affected transactions
Lot / tracking attributeTrack-by capture, labels and receipt historyInventory reviewer; attribute correction under authority
StatusInspection, damage and quarantine recordsQuality / inventory lead; status action, not automatic sellable stock
Timing / duplicate recordPhysical event time, posting time and linked task IDsProcess owner; correct once and preserve the first count

Choose the correction that matches the cause. A missed transfer needs a supported movement correction. A wrong pack conversion needs master-data review and a check of affected transactions. A real shortage may require a quantity adjustment. A damaged item may require a status change without changing total physical quantity. These actions can have different permissions and downstream effects; calling all of them an inventory adjustment hides those differences.

Keep before and after quantities, unit, reason, linked evidence, executor, reviewer and timestamps for an approved adjustment. Obtain client confirmation when the contract or authority matrix requires it. Do not create a blanket rule that every minor discrepancy waits for customer approval, and do not treat an internal tolerance as authority to change another owner's balance. Retain small recurring differences for trend review even if they use a streamlined approval path.

Extensiv's June 2026 audit procedure describes reconciliation that can move or adjust inventory and cannot be undone. It also says allocated inventory must be unallocated before that job can reconcile. This is a product-specific dependency, not permission to drop order reservations casually. An authorized operator should first plan the affected orders and verify the actual workflow. Extensiv audit procedure

Release only the stock and orders that pass

When verified stock is in the wrong place for picking, the next decision is an internal move. Use the 3PL pick-face replenishment SOP to check existing tasks, move quantities, destination capacity and the handoff window.

Confirm identity and quantity, an explained or authorized discrepancy outcome, fulfillment-eligible status, rechecked order allocation and correct removal of the relevant hold. A location count does not make quarantine stock sellable. If another verified location has the same owner's eligible item, assess using it rather than stopping every order in the building. Keep the unresolved stock distinct and record which orders remain affected.

Before resuming work, the outbound lead acknowledges the revised allocation and the floor team receives the release instruction. Check already-created pick tasks for obsolete quantities or locations. The inventory lead closes the count only after the movement control is correctly removed and any remaining exception has an owner. A new task is safer than silently leaving a picker with instructions generated from a balance that no longer exists.

When a pause ends, review the queue of work held during it. Release current instructions in a controlled order and confirm the first affected transaction uses the corrected balance and location. If a correction fails to post or a hold remains in another interface, keep that inventory exception open. The count record should distinguish physical verification, approval, successful system recording and release, since each can finish at a different time.

Suggested business flow: verify identity and units, reconcile movements, independently recount unexplained differences, authorize the right action, then recheck orders and release conditions.
Suggested operating flow, not a software state machine. Release only verified, eligible stock and orders.

6. Four hypothetical cases with different outcomes

A. Six units left before the system update

The scope is one original bin. At 10:00 its snapshot shows 40 units. A pick record proves six units physically left that bin at 10:05. The counter finds 34 at 10:12; the movement posts at 10:14. Check the same owner, SKU, unit and source, plus evidence that the picked six belong to that event. Look for other movements in the interval rather than assuming this single record explains everything.

With no other relevant movement, expected stock at 10:12 is 40 − 6 = 34; physical stock is 34, so the aligned variance is zero. Resolve the recording delay and preserve the first observation. Do not deduct another six. Changing the system from 40 to 34 before the pending pick posts could leave a recorded 28 against a physical 34. This is the counterexample to changing a balance immediately to match a count.

Original-bin timeline: system snapshot 40 at 10:00; six units physically picked at 10:05; 34 counted at 10:12; the same pick posts at 10:14. Aligned expectation is 34 and variance is zero.
Hypothetical case with no other relevant movement. A physical move and its later posting are one event, not two deductions.

This scenario represents insufficiently controlled movement, not a claim about how every WMS posts picks. Reopen the affected work only after confirming the pick posted once and allocations still make sense. The prevention is a documented count cutoff, a known treatment for pending picks and a dependable handoff between inventory control and outbound. The batch picking guide covers maintaining item and order identity after collection.

B. Twenty units total still hides two owner variances

Two clients use the same SKU text. Client A expects 10 units and the owner-specific check produces eight; client B expects 10 and the check produces 12. The combined expected and observed totals are both 20. Yet the apparent owner variances are −2 and +2. Review client identifiers, receiving references, item labels and transfer history. A shared SKU label alone cannot establish ownership of the disputed units.

Client A expects 10 and counts eight; client B expects 10 and counts 12. Total 20 equals 20, but owner variances minus two and plus two cannot be offset.
Hypothetical owner-level comparison. Confirm attribution from evidence and keep uncertain goods unresolved.

Keep the attribution issue open until the evidence identifies the goods and the needed correction. If identity cannot be established, use pending-identification status rather than assigning the extra two to the client who is short. Even when the physical owner is verified, a surplus does not prove its source. The outcome is an ownership investigation, not a cross-client offset. Prevent recurrence by validating owner identity at receiving and transfer; use the receiving SOP for that separate entry checkpoint.

C. Two cartons are not automatically two units

The system has 24 eaches; the counter sees two cartons. Confirm the item and owner, the approved 12-each pack specification and whether these are intact full cartons or mixed remainders. Printing on a carton is supporting information, not unconditional proof of contents. In this hypothetical case the approved method permits opening both, and the counter verifies 12 units in each. The comparable physical total is 2 × 12 = 24 eaches. Extensiv's UoM overview distinguishes primary and packaging units in 3PL Warehouse Manager; verify the actual item definitions. Extensiv units guide

The result needs no quantity reduction. Preserve the observed two cartons, the conversion and the verified content count. If the pack definition is wrong, send that field for master-data review rather than applying a one-off conversion invisibly. If a carton cannot be checked under the agreed method, mark it unresolved. Do not count it as empty or certified full. Improve the worksheet so carton observations and the primary counting unit are both visible.

D. A confirmed shortage changes order release

A location is expected to hold 15 eaches. Two independent observations confirm 11, and the team rules out identity, units, status, other positions and relevant movements. There is no verified eligible alternative stock. Two affected orders require eight and six units, totaling 14. The unexplained residual shortage is four units, and available physical stock cannot support both complete orders.

An authorized operator records the supported 15-to-11 quantity action. The outbound lead rechecks reservations, releases the eight-unit order if its other checks pass, and holds the six-unit order pending replenishment or an approved split. Reserving eight of the 11 leaves only three uncommitted units; they do not complete the second order. Preserve the customer decision and the remaining requirement if partial shipment is later approved. Investigate the recurring cause, but do not keep obsolete allocations active while waiting for a perfect explanation.

7. Measure first results without hiding the work

Define first-count record accuracy as valid records with no variance on the first comparable check divided by valid records with a completed first check. The comparison must include owner, location, unit and required attributes. A matching grand total cannot erase owner or lot errors. Preserve the first result after corrections; an adjusted balance agreeing with the shelf is an outcome, not a replacement for the original accuracy observation.

Define invalid independently from inaccurate. A valid observation can show a genuine quantity or lot discrepancy and belongs in the first-result measure. An invalid comparison may lack a usable cutoff or a verified unit and requires correction or replacement. Keep the reason, original observation and linked follow-up. Review exclusion decisions with the inventory lead, so a difficult difference does not disappear simply because someone labels the task invalid.

For a hypothetical assigned population of 115 record tasks, 100 have valid completed first checks: 92 match and eight differ. Ten are unfinished and five are invalid pending correction. First-count accuracy is 92 ÷ 100 = 92%. First-check plan completion is 100 ÷ 115, about 87%. Show the ten unfinished and five invalid separately, keep their follow-up visible, and do not remove problem records merely to improve the denominator.

Track due and overdue planned work, discrepancy closure time, recurring causes and orders affected by inventory problems. Define closure as the required resolution and release review, not merely typing a new quantity. Report open-case age alongside closed-case duration, so old unresolved work remains visible. A task can have a completed first check and an open discrepancy. Keep those two measures separate when discussing the backlog.

Report unique coverage as well as repeated observations. Review repeated differences by owner, location, unit and process, without netting gains against losses to produce an accuracy score. High-risk targeted checks are a selected population; their result is not a warehouse-wide random-sample estimate. For incident ownership across the inventory, shipping-software and carrier boundaries, use the shipping diagnosis framework.

8. Start with reusable records and a limited pilot

Use the record below as a shared reference between the count team, inventory reviewer and client contact. Keep the first count, recount, movement evidence and final decision in separate fields. Link replacements to invalid tasks instead of deleting their history. The accompanying release checklist names affected stock and orders so the count can finish without hiding unfinished fulfillment work.

Record blockFields to keep
Task identityTask ID, client, warehouse, location, SKU, lot/serial/other attributes, status, unit
Baseline and observationsTime basis, expected quantity and report reference, first-count quantity/time, recount quantity/time
Movement and investigationMovement IDs, physical/posting times, quantity and unit, source/destination, reason and evidence
Decision and closureAction record, before/after, authority, executor, reviewer, affected orders, release check, close time
Discrepancy and release checkComplete / hold evidence
Confirm scopeOwner, item, all selected positions, attributes, unit and common time basis
Preserve evidenceFirst count, independent recount, baseline, movement and content checks
Identify impactAffected inventory, orders, reservations and released pick tasks
Approve the actual actionRecord correction, master data, quantity or status; authority and client confirmation if required
Verify recoveryIdentity/quantity checked; outcome authorized; status eligible; allocation rechecked; holds correctly removed
Close or hand offReviewer, close time, unresolved work, next owner and next check

Pilot one client or a limited physical area. Define scope and authority first, then observe actual task time and discrepancy types. Review units, owner labels and delayed movements with the people doing the work. Repair those causes, rerun a controlled count where necessary and check order release before expanding. A trial period is a planning choice, not a promise to reach a particular accuracy in a fixed number of days.

Expand only when another shift can follow the same record and resolve an exception without guessing. Review due coverage, review capacity and customer commitments together. Keep room for urgent counts alongside the schedule. Where differences recur, follow the specific process that created them rather than increasing count frequency alone. Counting can reveal an error repeatedly while an unchanged receiving, transfer or unit process keeps recreating it.

Once inventory and orders are confirmed releasable, use actual package data to check available transportation services, rates, labels and tracking through EasyShippingX's enabled channels. The inventory controls described here belong in the warehouse's operating process and appropriate systems. Finish the physical shipment with a documented carrier handoff, so verified stock becomes an identifiable parcel with evidence of custody transfer.

Warehouse cycle counting FAQ

It can support ongoing verification, but it does not automatically replace counts required by a client, finance team or auditor. Confirm those requirements separately.

Set frequency from risk, transaction activity, recurring errors, commitments and available capacity. The 12/4/1 annual frequencies in this guide are planning assumptions.

Other controlled operations may continue. For the selected stock, use a local pause or a verified movement-reconciliation process. Check your actual WMS and existing task behavior.

No. Keep it as not counted until an observation is completed. An empty location needs an explicit physical check.

No. Timing, location, unit, owner or status issues can require different actions. Preserve the evidence and use the applicable approval authority.

Inspection or damage status, reservations and order rules can still prevent release. Verify eligibility and allocation before removing the relevant hold.

Sources and methodology

Sources reviewed November 8, 2026. Vendor features apply to the named product and documentation. The SOP, classifications and release controls are suggested operating procedures. Workloads, timestamps, case quantities and metrics are hypothetical examples.