Omnichannel retail is usually described as a customer experience strategy. Operationally it is an inventory accuracy strategy with a marketing cover. Every promise made online is a promise about a physical unit in a specific place, and the system either knows that or it does not.

The failure is the last unit, not the last mile

Delivery logistics are solved; unit-level truth is not.

A promise to deliver in two days is easy; a promise to have it on the shelf in an hour is hard.
The failure mode is not a late truck, it is a "phantom unit" that existed only in the database.
One phantom unit cancels one order and creates one disappointed customer who tells other people.
The cost of a cancellation is not the lost margin; it is the pickup trip and the trust.
Accuracy at 95 percent sounds good and means one order in twenty is at risk.

Accuracy is a number with a timestamp

An accuracy figure without a date is a memory, not a measurement.

Quote accuracy as "percentage of SKUs with zero variance at a stated time", not as a vague average.
A store that was 98 percent accurate last quarter may be 88 percent after a busy weekend.
Unit accuracy and value accuracy differ, and the customer promise depends on unit accuracy.
Small-ticket, high-count items are where phantom units accumulate.
Report accuracy per location and per category, because a single blended number hides the failures.

BOPIS: the promise made at checkout

Buy-online-pick-up-in-store is a three-part commitment, and each part can fail.

Availability at the moment of the order: the unit must exist and be findable.
Pickability within the promised window: staff must locate it before the customer arrives.
Handover at the desk: the correct order must reach the correct person.
The most common failure is the first, and it is invisible until someone walks in.
Reserve logic must hold the unit against other orders, or two customers are promised the same item.

The store as a fulfilment node

Once a store ships orders, its inventory stops being a display and becomes a warehouse.

Picking paths, pack stations and carrier cut-off times are warehouse concepts that arrive unannounced.
Store staff have customer-facing duties, so picking must fit between them.
A store that cannot find an item in five minutes should not be in the ship-from-store network.
Reserve stock in the back room needs its own accuracy discipline, separate from the sales floor.
Capacity limits should be explicit; a store shipping forty orders a day needs space, not goodwill.

The reserve stock problem

Back rooms are where accuracy goes to die quietly.

Reserve stock is high-density, low-visibility and rarely counted.
Cartons are received, stacked and recorded by quantity, not by unit.
An item that is in the back room but recorded as on the floor is effectively lost.
RFID read points at the back-room door catch the movement between zones.
Count the reserve separately, and reconcile it against the floor before trusting total availability.

Ship-from-store and the pick path

The pick is the moment where inventory truth becomes a customer outcome.

The pick list should be ordered by physical location, not by order line sequence.
A missing item must trigger an immediate, recorded exception, not a silent substitution.
Partial shipments damage trust more than a delayed shipment, so decide the policy in advance.
Pick confirmation should remove the unit from availability at once, not at dispatch.
Measure pick time per order and cancellation reason per store; they predict customer experience.

Cycle counting without stopping the store

Continuous counting is the only way to keep accuracy high in a live store.

Count a small rotation daily rather than a full count quarterly.
Count by exception: high-value, high-velocity, and items with recent negative adjustments.
Count where the reader is, so counting does not require closing the department.
Compare counted against recorded, and investigate the variance, not just correct it.
Track the trend of accuracy; a slow decline is easier to fix than a sudden collapse.

Returns and the reverse flow

Returns are inventory events, and treating them as paperwork is the classic mistake.

A returned item must be inspected, restocked and made available again, or written off.
Restocking lag is invisible in the system and very visible to the next customer.
Items returned to the wrong store distort both stores' availability.
Reverse logistics without RFID creates a class of items that exist in neither location.
Publish the restock lag as a metric; it is usually measured in days and should be hours.

What Odoo needs to hold

The ERP must express availability as a promise, not as a count.

Stock by location, including floor, back room, in transit and reserved.
Reservation records with an owner and an expiry, so a stuck reservation does not block a sale.
An availability service that answers "can this store fulfil this order now" quickly.
Item-level serial or EPC tracking where the promise depends on a specific unit.
An audit trail for every adjustment, because accuracy debates are settled by history.

The exception is the product

Handle the exceptions deliberately and the promise survives.

A not-found item should generate a task with an owner and a deadline, not a status change.
The customer should be told quickly, with an alternative, rather than late with an apology.
Repeated not-found at the same location is a process defect, not bad luck.
Exception rates by store, by shift and by category tell you where to invest.
A visible exception log changes behaviour faster than any training programme.