The demand your order data can't show you

Your reports are built from things that happened. Orders, revenue, units, conversion. All of it is a record of demand you met.

The demand you missed doesn't appear anywhere. Someone searched, found nothing, and left. There's no row for that.

The signal nobody reads

Forty people searched for a linen shirt last week. You don't stock linen shirts. In your analytics, that week looks fine — your conversion rate is calculated over the people who found something, and the forty aren't in the numerator or the denominator of anything you look at.

Your merchandiser's summary reads "shirts performed well". Nobody mentions the shirt that would have.

Zero-result searches are the cleanest demand signal in a store, and most teams never see them. They're sitting in the search logs, uncounted, because nothing downstream is built to ask.

Why forecasts miss it too

Standard demand forecasting extrapolates from sales history. Sold twelve last month, similar this month, reorder accordingly.

That works if you already stock the thing. It's structurally incapable of telling you about the thing you don't, because there's no history for a product you never listed. The forecast is a mirror of your catalog, not your market.

It also needs history to exist at all. Ask any tool to forecast for a store that's been live six months with a few hundred orders and it will politely produce noise. Which is exactly the store that most needs the help — you're deciding what to stock precisely when you have the least evidence.

Leading indicators exist and you already have them

Orders are a lagging indicator: the end of a decision made earlier. The earlier parts are in your store already, and they're more useful:

• Searches that returned nothing. Explicit, articulated demand for something you don't have. As direct as it gets.
• Add-to-carts that never converted. Intent that hit a wall — price, shipping, a variant out of stock.
• Repeat searches for a variant. Ten people looking for size 12 last week is a restock signal a fortnight before it's a stockout.
• Velocity per variant, not per product. The classic apparel trap: the product looks healthy, the M sold out on Tuesday, and the XS is dead stock propping up the average.

None of this needs new tracking. It's in the store — it's just not wired to anything that acts on it.

What we do with it

Our inventory agent blends actual sales velocity with those leading indicators, which is what lets it say something useful about a young store rather than shrugging until there's a year of history.

It projects stockouts from live velocity, surfaces the searches that returned nothing, and drafts a restock order — supplier, variants, quantities — for someone to approve. Not place. Approve. The agent proposes; a person decides and their name goes on the record.

The unmet-demand list is the part merchandisers actually want. Projected stockouts are useful and they're a known problem. "Forty people asked for this and we don't sell it" is a decision nobody in the building currently has the information to make.

Do this without buying anything

Pull your zero-result searches for the last 30 days, group them, sort by count. Most search tools have this in an export; if yours doesn't, it's a query.

Read the top twenty. Some are typos. Some are things you'll never stock. And some — usually more than people expect — are a list of products your customers told you they wanted, in their own words, on a specific date, and nobody was listening.

That list costs you an afternoon and it's the highest-signal document in the store.