How to forecast inventory with no sales history

The blind spot, stated plainly

Forecasting apps learn seasonality from your own sales. That is a sound approach once you have years of data, and a useless one before you do. Read the requirements each app publishes and the pattern is consistent — these were taken from the App Store listings on 1 August 2026:

None of this is a flaw in those products. It is a consequence of the method. But it does mean that if you are opening a store, launching a product, or heading into a season you have not traded through yet, the tool goes quiet exactly when you need a number.

And this is not only a new-store problem. Apparel and DTC brands launch products continuously; last year’s data does not transfer to this year’s styles. A mature store can be in a cold start on half its catalogue.

Two things that exist before the first sale

You cannot conjure the missing history. You can use information that does not depend on it.

1. The calendar of the country you sell in

Retail peaks are not discovered — they are known. Black Friday falls the day after American Thanksgiving. The French soldes run twice a year. Mothering Sunday in the UK is in March, not May. Japanese gift seasons, Ochugen and Oseibo, land in July and December. A statistical model has to survive one of these before it can see it; a calendar knows it in advance.

UreyukiBox applies a fixed calendar coefficient based on your store’s own country — Japan (13 events), Canada (10), the UK and France (9 each), the US and Australia (8 each), Germany (7) — each cut in that country’s time zone. Countries without a calendar get no adjustment rather than a borrowed one.

2. How comparable products in your catalogue started

Your new product has no history, but products like it do. UreyukiBox looks at how similar items sold in their first thirty days on sale and turns that distribution into an opening-order range — conservative, standard and aggressive. It looks for comparables in a fixed order: same product type first, then same vendor, then the store as a whole, stopping at the first level with real data.

It also reports how many comparable products it found, so a suggestion built on two items does not read the same as one built on forty. And it never overwrites your reorder points — it is a number to look at while you decide.

Refusing to answer is a feature

The failure mode worth avoiding is not a missing number. It is a confident wrong one, because a fabricated figure gets ordered against and paid for.

• No comparable products? The opening-order estimate returns nothing instead of a default.
• Not enough data to measure variability? Safety stock falls back to the plain deterministic reorder point rather than inventing a spread.
• A product that sells rarely and irregularly? A moving average lurches every time one unit enters or leaves the window, so intermittent demand is switched to a method built for it (Croston / SBA) automatically.
• A product listed twelve days ago? Its average is divided by the days it has actually been on sale, not by a flat thirty — otherwise every new product looks like it is barely selling.

One thing no app can honestly claim to solve: days when you were out of stock look identical to days with no demand, because Shopify does not retain inventory history. We do not claim to correct for that, and we do not publish accuracy percentages.

What to do this week

1. 1. List the products that will carry your next season but have under three months of sales. That set is your exposure.
2. 2. For each, find two or three comparable products you already sell and look at their first month, not their steady state.
3. 3. Mark the calendar peaks for the country you sell into, and work backwards through your supplier lead time to get the order date — not the sale date.
4. 4. Write down what you assumed. When the season is over, that note is the only thing that tells you whether you were wrong for the right reasons.