Stock optimisation with AI, including on Shopify

To optimise stock, AI does four things: it forecasts demand product by product and day by day, it decides how much to order, move or mark down by weighing the risks, it explains every decision, and it then measures what the decision earned. It does not replace your knowledge of the trade. It does, across hundreds of products, the calculations nobody has time to do by hand.

That is what Sway One does for shops that sell in store and online, Shopify included. Here is what AI really brings, its limits, the seven-step method, and the specific case of Shopify.

Sway One and Sweet 1.4: the app and its AI

Sway One is the app that connects to everything: e-shops like Shopify, PrestaShop, WooCommerce or Magento, tills like Square, SumUp, Lightspeed, Zettle or Hiboutik, ERPs like Odoo or Dolibarr, logistics providers like Bigblue or Amazon FBA. More than twenty tools, which you connect yourself from the Connections page, in a few minutes.

Sweet 1.4 is its decision AI. It forecasts, suggests restocks, transfers and discounts, and no longer discounts by reflex. Every recommendation arrives in English, with its reason, its estimated impact and a confidence level: you approve it in one click, adjust it or discard it, for all your shops from a single screen.

What AI does better than a spreadsheet

A spreadsheet applies one rule. AI chooses the rule product by product, and adjusts it with what is going on around you.

Forecast day by day, product by product

An average of last month's sales misses weekends, public holidays and the start of the sales. A day-by-day forecast sees them.

You still need to know when to trust it. Horizon, the forecast of the Sweet family, pits a statistical model against a day-by-day forecast for each product, replays them over the last 28 days, and weights them according to which one predicted best. On the difficult products of our benchmark, the error is cut by up to more than half compared with the day-by-day forecast alone. The details are in Horizon, product-by-product stock forecasting.

With Sweet 1.4: the forecast is shown on the product page, with the horizon of your choice and its uncertainty band.

Read signals from the outside world

Rain, a cold snap, a topic rising in searches: demand moves before your sales show it. A spreadsheet does not see it.

With Sway One: your sales are cross-checked with the upcoming weather, search trends and the news. The impact of each signal is calibrated on your own sales, family by family, and its reliability is shown. A sourced recap of the news in your sector is produced every day.

Decide under uncertainty

A forecast is never exact. The real question is: how much should you order, knowing that missing one piece and having one too many do not cost the same?

This is the critical fractile calculation, done for each product with your margin, your holding cost and the salvage value of unsold items. Then comes the arbitration: among hundreds of possible actions, which ones are worth it? Sweet Instinct rereads the actions that the filters discarded and estimates the probability that each one is worth it after all. On eight demo shops, it finds 2.2 times more profitable ones than sorting by estimated gain: see Sweet Instinct, the AI that rescues good actions.

With Sweet 1.4: the critical fractile and Sweet Instinct are part of every decision tree, and Sweet Instinct only receives figures, the category and the brand: no product name, no reference.

Only mark down if it pays off

A conventional tool puts on sale whatever has not moved for X days. But a slow item is not a dead item.

With Sweet 1.4: a promotion or markdown is only suggested if it earns more than waiting at full price, and never below the purchase price outside the sales. This is the heart of Sweet 1.4.

The limits of AI for managing stock

Dirty data gives wrong forecasts

An AI learns from your history. If stock-outs show up in it as days without demand, if the same reference exists under two codes, it forecasts wrongly with confidence.

With Sway One: products are matched across channels by reference or barcode, and stock-out days are corrected before forecasting.

New products have no history

No model can forecast a product launched yesterday from its own sales.

With Sweet 1.4: a product with fewer than three days of sales borrows the rhythm of the five closest products in its family, then moves away from it as its own sales come in.

AI does not know everything

It does not know, for example, how strongly your customers react to a price increase. Rather than gamble, Sweet 1.4 only suggests an increase if a signal justifies it: weather, trend or season. Language models, for their part, provide quantified hypotheses and explanations; the calculation decides.

An AI has to be measured

A convincing demo proves nothing. An inventory management AI is judged on sales it has never seen, against a baseline. That is what our benchmark does, and what Sway One then does for you, decision by decision.

How to optimise my stock with AI in 7 steps

1. Connect all your sales. Tills, e-shop, logistics provider, ERP: an AI that only sees one channel forecasts only one channel. With Sway One: more than twenty connections, read-only.
2. Clean up the history. Unique references, barcodes, stock-outs corrected. With Sway One: matching by reference or barcode, stock-outs corrected.
3. Set your business rules. Service level, supplier lead times, thresholds per family or per product, minimums and free-shipping thresholds. With Sway One: thresholds per family and per reference, supplier constraints in your deployment's configuration.
4. Let the AI forecast. Product by product, day by day, with a range rather than a single figure. With Sweet 1.4: forecast with an adjustable horizon and an uncertainty band.
5. Add external signals. Weather, search trends, news, the calendar of sales and public holidays. With Sway One: signals calibrated on your sales, family by family.
6. Approve quantified decisions. Each recommendation with its reason, its estimated gain and its confidence. With Sweet 1.4: restocks, transfers, discounts and pre-orders ranked by impact; optional autopilot, above a confidence threshold set with you.
7. Measure the outcome. Compare what each decision was meant to earn with what it earned, and recalibrate. With Sway One: actual impact measured after the horizon, following estimates recalibrated.

AI and stock on Shopify

What Shopify does natively

Shopify tracks the stock of each variant by location, whether a shop, a warehouse or a logistics provider, and offers sales and stock reports. If all your shops take payments with Shopify, your sales are already in one place.

What is missing is the decision: how much to reorder, when, where, and at what price.

What forecasting apps bring

The Shopify App Store has many stock forecasting apps. According to their listings, most forecast demand per reference and prepare supplier orders. That is already a big step up from the spreadsheet.

Their typical limits come from their starting point: they analyse the sales of the Shopify site. If your shops take payments on another till, those sales escape them. Multi-shop often boils down to stock per location, with no transfer suggested between shops. And we did not find weather and trend signals, markdowns placed within the statutory sales periods, or measurement of the real impact of each decision advertised on their listings.

What Sway One does on Shopify

Sway One connects to Shopify in a few clicks and reads your products, variants, orders and stock. Then it fills in what remains open:

Your ERP or your logistics provider are added alongside: Odoo, Dolibarr, Bigblue, ShipBob, byrd, Amazon FBA. And your AI assistant can read and approve recommendations via MCP.

The measured results of Sweet 1.4

Before shipping Sweet 1.4, we replayed it on four simulated shops, built from the public catalogues of fifteen retailers: sport, outdoor, furniture, fashion. Twenty-six weeks, from March to August 2026, with the same sales for each version.

The reference is the retailer's routine: one review a week, restocking based on the last 60 days of sales. Sweet 1.4 is added on top, with 10 recommendations applied per day. The margin is counted after a carrying cost of 25% a year.

These are simulated shops, with estimated margins and costless transfers: these are orders of magnitude, not a promise. The real gain is measured on your data.

The checklist for choosing AI inventory management software

• It reads all your sales and all your stock: e-shop, tills, ERP, logistics provider.
• It forecasts by product and day by day, with a range.
• It corrects stock-outs in the history and handles new products.
• It cross-checks your sales with weather, trends and the news, showing the reliability of each signal.
• It decides in euros: margin, holding cost, stock-out risk.
• It only marks down if it pays off.
• It reasons shop by shop and suggests transfers.
• It explains every recommendation, with a quantified impact and a confidence level.
• It measures the outcome and recalibrates itself.
• It has been measured against a baseline before it reaches you.

This list is the specification of Sway One and Sweet 1.4: Sway One ticks every box.

Why Sway One and Sweet 1.4

• An app that plugs into everything: Shopify and more than twenty tills, e-shops, ERPs and logistics providers, connected by you.
• All your shops and your e-shop on one screen, filterable shop by shop.
• A measured decision AI: Sweet 1.4 forecasts day by day, reads the weather, trends and the news, and no longer discounts.
• Clear recommendations, in English, approved in one click.
• From restocking to the purchase order, with minimums, free-shipping thresholds and lead times.
• Every decision measured, and your AI assistant connected via MCP.

On four simulated shops, Sweet 1.4 adds +4.7% to +11.6% of margin after carrying cost to the retailer's routine, with fewer stock-outs in all four cases.

Sway One costs €289 excl. VAT per month, with a deployment dedicated to your shop. To find out what it would earn you, the gains calculator puts a figure on three scenarios over twelve months, subscription deducted. To see it on your data, book a demo.

For the basics without AI (indicators, reorder point, ABC method), read Stock optimisation: indicators, methods and software.

Frequently asked questions

How do I optimise my stock with AI?

Connect all your sales, clean up the history, set your supplier rules, then let the AI forecast day by day and suggest quantified decisions. Approve the ones that suit you and measure what they earned.

Which software for stock optimisation with AI on Shopify?

Sway One, with its decision AI Sweet 1.4. It reads Shopify, but also your tills, your ERP and your logistics provider, reasons shop by shop and measures the impact of each decision.

Can AI forecast the sales of a new product?

Not from its own sales alone. Sweet 1.4 borrows the rhythm of the closest products in the same family, then relies on the product's own sales as soon as they come in.

Is a Shopify app enough to manage the stock of my shops?

If all your shops take payments with Shopify and you do not need transfers, it is a good start. With another till or several shops to balance, you need a tool that reads all your sales.

Is my data used to train the AI?

No. Each customer has their own deployment, and your data is not used to train any model. For Sweet Instinct's decision, only figures, the category and the brand are sent.