Stock counts and checks a person must still do

A forecast starts from the stock number in your system. If that number is wrong, every suggestion built on it is wrong too. A person has to confirm it against the shelf.

Count a few items often

You don't have to close the shop to count everything. A cycle count is an auditing procedure that involves counting a small, specific subset of inventory in a continuous, regularly repeated sequence.¹ It's less disruptive than stopping to count all items at once.

Which items come first? Wikipedia describes one method where items with a higher determined value are counted more often, while items that have little movement are seldom counted.¹ Another counts at the moment stock hits a trigger: the inventory is audited at re-order points, when a certain amount is used, or when inventory levels are low.¹

That second method suits a small shop well. When the reorder point comes up, count the shelf before you order.

Why a person has to do it

The alternative is trusting the system. Out-of-stocks can be found from inventory data, but only depending on its accuracy.² Counting by hand has its own weakness: how well manual audits work depends heavily on their frequency and timing, and on avoiding human counting errors.² And cycle counts, like full counts, can introduce errors if the process is poorly executed.¹

So the job isn't "count once and relax." It's a routine, done the same way each time.

The checks

Each of these is a thing a forecast can't see.

• Shelf and back room. Count what's on the shelf, in the back and on order but not yet received. Write down where each lot is, so one count doesn't miss the other.
• Damaged, expired or returned goods. Stock that can't be sold still sits in the system as stock. Remove it from the sellable count and note why.
• Missing items. If the count is lower than the record and nothing was sold, look for the cause. Check for wrong deliveries, unrecorded breakage and items moved to another spot. Don't just overwrite the number and move on.
• Delivery checks. Compare what arrived with the delivery note and the order. A short delivery that goes unrecorded makes the system think you have more than you do.
• Odd sales. A very large single sale, a sale at a big discount or a refund can distort a product's history. Statisticians call such a point an outlier, one that sits well away from the rest.³ Mark those weeks so the tool doesn't read them as normal demand.
• Moved or mislabelled stock. The same item in two places, or in the wrong box, is easy to miss in a spreadsheet.

Keep a record

Write down the date, who counted, what was found and what was corrected. A short log is enough. Next time a suggestion looks off, the log shows whether the count was fresh.

If a count turns up a large gap, pause the order for that product until you understand it. Ordering from a wrong number locks in the error.

Where AI fits

An AI tool can help with the paperwork around counts. It can draft a count sheet, sort the list by value or suggest which items to count next. It shouldn't be the one deciding what's on the shelf. It has no eyes in the stockroom.

Think of the forecast and the count as two halves of the same job. Problems in the data are listed under common causes of bad suggestions. To keep records tidy between counts, start with the list itself.