AI for Retail Counts the Queue. The Money Dies in the Stockroom

At half past five on a Sunday, Renee is in the back room of her clothing and footwear store in Ballarat, halfway through a stocktake that keeps telling her a story she does not want to hear.

Her technology has had an excellent week. The webchat on her site answered every after-hours question about opening times and sizing. The people counter graphed foot traffic by the hour. The social scheduler posted the new season drop at exactly the right time on Thursday night.

None of it noticed that her best boot broke in size nine on the first Tuesday of the season, and stayed broken for six weeks until the sales rep happened to call in. None of it counted the women who asked for a nine, got a shrug, and bought the same boot somewhere else that afternoon. And none of it has ever mentioned the knitwear rack, untouched since July, still at full price, waiting to become January's fifty per cent problem.

A shop records every sale it makes. It never records the sale it turned away.

Nobody decided to run out. The shop simply watches the register, never the shelf. That is the gap in how AI for retail is being sold: chatbots, personalisation engines, people counters, all of it aimed at the customer who is already in the shop. I build these systems for Australian businesses, so let me say the uncomfortable half out loud. Most small retailers are not short of customers. They are short of the exact size the customer walked in holding money for.

What AI for retail gets right

Give the front-of-shop tools their due, because the case for them is real.

A question about sizing or stock that goes unanswered at nine at night is a customer teaching themselves to buy from the marketplace instead, and a chatbot that answers from your own product data is doing work nobody was going to do anyway. I made the broader case in AI chatbots for small business.

But notice what all of it touches: the person already looking at you. The industry builds there because a conversation is easy to demo, and nobody has ever filmed a reorder report for a launch video.

The half of AI for retail nobody demos

Now read the same week from the stockroom instead of the counter.

Your point of sale already holds, for every style, the rate of sale by size and colour, the stock on hand, and the weeks the supplier takes to deliver. Which means it holds a date. The date each winner will break in its best size, weeks before it happens.

When that date passes, nothing turns red. The missing sales are invisible because a customer who walks out was never recorded anywhere. A broken size does not look like a problem. It looks like a quiet week for that style, which reads as fading demand, which talks you out of the reorder you needed six weeks ago.

The same file holds the opposite number: the racks where nothing has moved for ninety days, the season's profit asleep on a hanger.

Your POS knows the day each bestseller will break. Nobody in the shop is paid to look up that date.

The sell-through maths nobody runs

Run your own shop as arithmetic.

Say a winning style sells six a week across its two best sizes, the reorder takes six weeks to land, and the break goes unnoticed for four of them. That is roughly twenty-four missed sales. At ninety dollars each, one style has quietly cost about two thousand dollars, and a shop with ten winners a year has lost the price of a good fit-out.

Both numbers are invented and yours to replace; the shape survives any honest figure.

Now the other end. Whatever your last stocktake showed sitting past ninety days, the markdown you will eventually swallow to move it is the rent that stock has been charging. Nobody clears every slow line, and nobody should chase a hundred per cent sell-through. But catching even half the breaks and half the sleepers is bought entirely with data already in the till.

Start with the break date, not the chatbot

Pick one workflow, not the whole shop.

A system reads the sales history you already hold, whether that lives in Lightspeed, Square or Shopify, and watches rate of sale against stock on hand and supplier lead time. When a winner is on course to break, it drafts the reorder while there is still time for it to land. You get it Monday morning for a yes or a no. Nothing is ordered on its own.

Then point the same loop at the other end of the shop. Every line past sixty days flagged, with a drafted mid-season offer while a small markdown still rescues the margin. The choice stays yours. The noticing stops depending on you.

That gives you two numbers each month: breaks caught, and dollars woken up before the sale rack.

Automate the counting, never the buying

Here is the line that keeps this safe.

No system chooses your range. The brands you back, the styles you punt on, the customer you dress: that judgement is the shop, and a model that reorders on arithmetic alone will happily fill your stockroom with last winter's winners. The system watches, counts and drafts. The owner reads, decides and signs the order.

Aim the software at the reorder date, never at the range.

What AI for retail costs

Work down this list in order and stop the moment something works.

• The tools already inside your POS. Reorder points, low-stock alerts, aged-stock reports. Usually half configured. Turn them on before you spend anything.
• A forecasting or inventory add-on, priced per store each month. Cheap enough to trial for one season and judge on one number: breaks caught.
• A custom build that reads your sales history and supplier lead times, drafts the reorders and the markdown lists, and writes back into the software you already run. A few thousand up to the mid teens of thousands, depending on how many systems it must talk to.

Hold that against one broken bestseller per season. What catches owners out is the running cost rather than the build, and I broke those layers apart in what AI actually costs a small business.

The Australian layer: the calendar and the customer list

Two things separate a system built for an Australian shop from an overseas template.

The first is the calendar. Most forecasting tools ship trained on a northern-hemisphere year, where Christmas is cold and the big sales follow winter. Yours is upside down: Christmas trade in summer, Boxing Day as the biggest day of the year, end of financial year in June. A tool that has never heard of EOFY will misread an Australian shop's whole rhythm.

The second is the customer list. Your loyalty file holds names, contact details and buying history, and Australian privacy law is tightening around exactly that kind of data. Ask where it is stored, whether it trains someone else's model, and whether you can take it with you when you change platforms. Those answers belong in writing before anything reads your till.

Common questions about AI for retail

What should a small retailer automate first?

The break-date report. It runs on sales data you already hold, and one season measures it with two numbers: reorders drafted in time, and sales saved.

Will AI replace retail staff?

No. The conversation on the floor and the taste in the buying are the product. The counting and the flagging around them are not.

Are AI chatbots worth it for a shop?

Sometimes, if after-hours questions genuinely go unanswered. Check that first. If your problem is broken sizes, a faster answer just tells the customer you are out of stock sooner.

Go back to Renee. She did not have a bad season. Her till knew the day the boot would break in size nine, and the week the knitwear stopped moving, and it was never asked to say either out loud.

If you want a straight read on what your stockroom gave away this year, that is what a first call is for. Book a strategy call and bring your last stocktake and your ten best sellers. I will count the broken sizes with you before we talk about building anything.