What AI can and cannot do with stock and reorders

An AI tool can read your sales history, spot a pattern and suggest what to order. It can't walk to the shelf and see what's really there. That split decides most of what follows. Keep it in mind.

What the tools do well

Forecasting from past sales is the main job. Software that predicts demand now often uses machine learning on historical sales data and outside factors.¹ A spreadsheet function can do a simpler version of the same thing. Microsoft says its forecasting function can be used to predict future sales, inventory requirements, or consumer trends.²

Practical jobs where a tool saves effort:

• Sorting a long product list so the slow movers and fast movers stand out.
• Drafting a first reorder list from recent sales, for you to edit.
• Explaining a formula you don't follow, in plain words.
• Turning a messy export into a tidy table.

None of these needs the tool to be right every time. They need you to check the result before acting on it.

Where it gets things wrong

An AI chat tool can give a confident answer that's false. People call that a hallucination, and Wikipedia describes it as false or misleading information presented as fact.³ In stock work that might be a made-up sales figure, a wrong total or a formula that looks fine and isn't.

Your own records can mislead it too. A forecast is only as good as the sales data fed to it. If an item sold out for two weeks, the sales dip looks like low demand. Wikipedia's page on stockouts notes that substitution disguises true demand, because shoppers buy something else when the first choice is gone.⁴ A tool reading that history may suggest ordering less of the very item you ran short of. See the page on forecast failures.

Counts are another weak point. The shelf count in your system may not match the shelf. Wikipedia says the system's figures are usable for spotting out-of-stocks only to the degree they're accurate.⁴ The tool can't tell you your numbers are off. It works with what you gave it.

What stays with a person

Four jobs stay with you, whatever software you use:

1. Count what's physically on hand. The counting page covers how.
2. Tell the tool about things it can't see, such as a supplier delay, a one-off bulk sale or a product you're dropping.
3. Compare the suggestion with the actual sales afterward. Forecast error is measured against actual sales.¹
4. Approve the order, and the money that goes with it.

The diagram shows the order of play.

A fair way to use it

Treat the output as a draft from a quick assistant who has never seen your shop. Ask for the working, not only the answer. If a suggested quantity has no reason attached, don't order from it.

Start with a few products you know well. If the tool's numbers look wrong for items you understand, they'll probably look wrong for the rest. Tidy the data first, using this data-prep page. Then learn the reorder arithmetic so you can check the tool.